Philip Kitcher The Advancement of Science (1993)
Philip Kitcher The Advancement of Science (1993)
Philip Kitcher The Advancement of Science (1993)
PHILIP KITCHER
Parts of Chapter 2 previously appeared in "Darwin's Achievement" (in Nicholas Rescher [ed.]
Reason and Rationality in Science, Washington, D.C.: University Press of America, 1975, 127-
190). Some short discussion in Chapters 3 and 4 are taken from "Theories, Theorists, and
Theoretical Change" (Philosophical Review, 87, 1978, 519-547). Chapter 8 contains sections
drawn from "The Division of Cognitive Labor" (Journal of Philosophy, 87, 1990, 5-21), and
from "Authority, Deference, and the Role of Individual Reason" (in Ernan McMullin [ed.] The
Social Dimension of Science, Notre Dame: University of Notre Dame Press, 1992, 247-270). I
am grateful to the editors of these books and journals for their kind permission to reprint this
material.
24689753
Printed in the United States of America
on acid-free paper
For Andrew and Charles
This page intentionally left blank
Preface
This is the book I have wanted to write ever since I began studying the history
and philosophy of science. More exactly, it is concerned with the topics that
have interested me the most throughout the past twenty-odd years. The ver-
sions of it that I would have produced at different times during that period
would have diverged, sometimes quite dramatically, from what I have now
written. I suspect, furthermore, that my ideas about the growth of scientific
knowledge will continue to evolve. These are not, I hope, my last thoughts
about the issues that occupy me. However, I also hope that their formulation
will help to stimulate improved thoughts, both for me and for others.
I owe an enormous intellectual debt to two people who taught me when
I was a graduate student at Princeton in the early 1970s: C. G. Hempel and
Thomas Kuhn. Although my treatment of virtually all the questions I address
differs from theirs, I could not have arrived at my own conclusions without
their deep insights. I have also been greatly influenced by the writings of
Alvin Goldman, Hilary Putnam, and W. V. Quine, all of whom have shaped
my ideas in important ways.
I began work on the writing of this book during 1988-89, when I had leave
from teaching at the University of California at San Diego. That leave was
largely made possible by a Fellowship from the John Simon Guggenheim
Foundation, and I am most grateful for the foundation's support. Earlier, the
development of my ideas had been aided by the opportunity to serve as co-
director of an institute to investigate new consensus in the philosophy of
science (held at the University of Minnesota), during which I had the chance
to listen to presentations made by many of the world's leading philosophers
of science. I would also like to thank the Office of Graduate Studies at the
University of California at San Diego for supporting a workshop on natu-
ralizing the philosophy of science.
Many people have read large parts of this book and offered valuable
comments on it. I am grateful to John Dupre and Elliott Sober for many
philosophical insights, and to Larry and Rachel Laudan for illuminating dis-
cussions. Michael Rothschild (of the UCSD Economics Department) provided
extensive comments on Chapter 8, which have enabled me to make consid-
erable improvements. From my colleagues Martin Rudwick and Robert West-
man I have learned much about the craft of history and about the more
viii Preface
specific aspects of the history of science for which they are well known. Martin
Rudwick's careful reading of Chapter 6 was particularly helpful. To the many
others who have listened to lectures in which I presented rough-hewn frag-
ments of the material of this book, and who have offered insightful sugges-
tions, my belated (and inadequate) thanks.
In recent years I have been fortunate to work with a group of extremely
talented graduate students. Their questions, comments, criticisms, and sug-
gestions have improved many parts of this book. In particular, I would like
to thank Gillian Barker, Michael Bishop, Jeffrey Brown, Sylvia Gulp, Michael
Dietrich, Bruce Glymour, Peter Godfrey-Smith, Gary Hardcastle, Todd
Jones, Alex Levine, Sam Mitchell, Eric Palmer, Neelam Sethi, and Andrew
Wayne, all of whom have left their mark on the pages that follow. I am also
grateful to Joe Ramsey for his preparation of the index.
During the past six years, my thinking about epistemology and the history
and philosophy of science has been greatly helped by discussions with Stephen
Stich and Steven Shapin. Neither is likely to agree with the conclusions of
this book, but both can pride themselves on having diverted me from even
sillier things that I might have said. I am also extremely grateful to ah anon-
ymous reader for Oxford University Press, who not only gave me numerous
specific suggestions for improvement but also convinced me that the bloated
penultimate version needed to be put on a diet.
Above all, I would like to thank three exceptionally perceptive philoso-
phers and generous friends, who sent me detailed written comments on the
entire penultimate draft. Small tokens of my thanks to Wesley Salmon, Kim
Sterelny, and John Worrall are scattered throughout the following pages, but
there is no way to convey in footnotes the more global suggestions they have
made that have helped me, again and again, to bring my ideas and arguments
into clearer focus. I am immensely grateful to them.
Finally, as always, Patricia Kitcher has provided sage advice, generous
encouragement, and warm support, without which this project would never
have been completed. Our two sons, Andrew and Charles, have offered both
distraction and delight. As I reflect on the writing of this book, I recall the
many baseball games, soccer practices, attacks on computer games, piano
festivals, tennis foursomes, and mountain hikes from which I have returned
to philosophy, refreshed. The joy of their conversation and their many en-
thusiasms deserve a better return, but this is the book I wrote and, with all
its flaws, it is theirs.
P. K.
La Jolla, Calif
June 1992
Contents
1. Legend's Legacy, 3
2. Darwin's Achievement, 11
3. The Microstructure of Scientific Change, 58
4. Varieties of Progress, 90
5. Realism and Scientific Progress, 127
6. Dissolving Rationality, 178
7. The Experimental Philosophy, 219
8. The Organization of Cognitive Labor, 303
Envoi, 390
Bibliography, 392
Index, 407
This page intentionally left blank
The Advancement of Science
My mistress' eyes are nothing like the sun;
Coral is far more red than her lips' red:
If snow be white, why then her breasts are dun;
If hairs be wires, black wires grow on her head.
I have seen roses damask'd, red and white,
But no such roses see I in her cheeks;
And in some perfumes is there more delight
Than in the breath that from my mistress reeks.
I love to hear her speak, yet well I know
That music hath a far more pleasing sound:
I grant I never saw a goddess go,
My mistress, when she walks, treads on the ground:
And yet, by heaven, I think my love as rare
As any she belied with false compare.
William Shakespeare
Sonnet CXXX
1
Legend's Legacy
Once, in those dear dead days, almost, but not quite beyond recall, there
was a view of science that commanded widespread popular and academic
assent. That view deserves a name. I shall call it "Legend."
Legend celebrated science. Depicting the sciences as directed at noble
goals, it maintained that those goals have been ever more successfully realized.
For explanations of the successes, we need look no further than the exemplary
intellectual and moral qualities of the heroes of Legend, the great contributors
to the great advances. Legend celebrated scientists, as well as science.
The noble goals of science have something to do with the attainment of
truth. Here, however, there were differences among the versions of Legend.
Some thought in ambitious terms: ultimately science aims at discovering the
truth, the whole truth, and nothing but the truth about the world. Others
preferred to be more modest, viewing science as directed at discovering truth
about those aspects of nature that impinge most directly upon us, those that
we can observe (and, perhaps, hope to control). On either construal, discovery
of truth was valued both for its own sake and for the power that discovery
would confer upon us.
According to Legend, science has been very successful in attaining these
goals. Successive generations of scientists have filled in more and more parts
Of the COMPLETE TRUE STORY OF THE WORLD (or, perhaps, Of the COMPLETE TRUE
STORY OF THE OBSERVABLE PART OF THE WORLD). Champions of Legend acknowl-
edged that there have been mistakes and false steps here and there, but they
saw an overall trend toward accumulation of truth, or, at the very least, of
better and better approximations to truth. Moreover, they offered an expla-
nation both for the occasional mistakes and for the dominant progressive
trend: scientists have achieved so much through the use of SCIENTIFIC METHOD,
Variants of Legend often disagreed, sometimes passionately, on details
of method, but all concurred on some essential points. There are objective
canons of evaluation of scientific claims; by and large, scientists (at least since
the seventeenth century) have been tacitly aware of these canons and have
applied them in assessing novel or controversial ideas; methodologists should
articulate the canons, thus helping to forestall possible misapplications and
to extend the scope of scientific method into areas where human inquiry
typically falters; in short, science is a "clearing of rationality in a jungle of
3
4 The Advancement of Science
muddle, prejudice, and superstition."1 Indeed, many advocates of Legend
would maintain that science is the pinnacle of human achievement not because
of its actual successes but in virtue of the fact that its practice, both in attaining
truth and in lapsing into error, is thoroughly informed by reason. Even those
whose historical situations lead into making mistakes still do the best they
can in the interests of truth, judging reasonably in the light of the available
evidence and bowing to whatever new findings expose their errors.
Who wereand arethese advocates of Legend? Many reflective people,
if asked for an assessment of science, would respond with a version of Legend.
Practising scientists have sometimes been more ambivalent, proclaiming Leg-
end on high days and holidays, sometimes unconsciously using it in formu-
lating their plans, sometimes, in their cups, confessing that the realities of
science do not reflect Legend's rosy glow.2 However, the most detailed ar-
ticulations of Legend have been provided not by the practitioners but by their
amanuenses in history of science, philosophy of science, and sociology of
science.
To find confident testimonials to the progressiveness of science and the
rationality of individual scientists, we need look no further than some widely
read writings of the 1940s and 1950s. James Bryant Conant's volumes of Case
Studies in Experimental Science, designed to convey to general readers "the
variety of methods by which [the physical] sciences have advanced" (Conant
1957 vii), attempted to recreate the evidence and reasoning that led scientists
to support novel ideas. Objectivity and rationality are the order of the day.
Similarly, Bernard Barber, reflecting on the cultural values of the "modern
world," recognized a "belief in and an approval of "progress" in this world,
a progress which is not necessarily of a unilinear evolutionary kind, but which
is somehow cumulative in the way in which science and rational knowledge
are cumulative" (Barber 1952 66; italics mine).3 Although the truth about
nature may be difficult to attain, Barber insisted that the cornerstone of
"scientific morality" is "devotion to the 'truth' " and that "it is worth endless
human striving to attain those provisional and approximate statements of
truth that make up the substance of science at any given historical moment
in its course of development" (Barber 1952 87).
Philosophical articulations of Legend are even more rich and impressive.
The failure of simple attempts to demarcate the cognitively meaningful (sci-
ence) from the cognitively meaningless (post-Kantian German metaphysics)
inspired the original members of the Vienna Circle and their philosophical
kin to investigate the characteristics of good science.4 Rudolf Carnap, Carl
1. The phrase is Bruno Latour's, who uses it for ironical effect. See (Latour 1988 6)
and also (Latour 1987 chapter 5).
2. James Watson's celebrated (1966) is a striking confessional. The reaction of many
reviewers is interesting and instructive. See, in particular, (Lewontin 1968).
3. Barber's reference to evolution appears to embody a common misunderstanding.
See (Williams 1966 34-55) and (Gould 1989).
4. In my judgment it is important to separate two movements with an overlapping cast
of characters. The earlier, logical positivism, took as its central philosophical problem that
Legend's Legacy 5
Gustav Hempel, Ernest Nagel, Hans Reichenbach, and Karl Popper endeav-
ored to analyze good science by focusing on questions about the confirmation
of hypotheses by evidence, the nature of scientific laws and scientific theories,
and the features of scientific explanation. They differed on many points of
detail, but some major articles of doctrine united them. Among these was
the conviction that the succession of theories in the physical sciences consti-
tuted a progression and that the achievements of earlier theories (confined,
in some versions, to the deliverance of statements about the observables)
were retained in later theories. Another was the understanding that, while
there is no systematic way to generate new hypotheses, once hypotheses have
been proposed there are principles for their proper assessment in light of
statements of evidence. Inspired by the work of Gottlob Frege and Bertrand
Russell, the architects of modern mathematical logic, logical empiricist phi-
losophers of science proposed to uncover the logic of confirmation, the logical
structure of theories and the logic of explanation, thus formulating with pre-
cision those canons and criteria that they took to be tacitly employed by
scientists in their everyday work. References to logic reverberate like drum-
rolls through the classic works of logical empiricist philosophy of science,
works that, because of their clarity, rigor, and attention to a wide range of
considerations, belong among the greatest accomplishments of philosophy in
our century.
So much for the dear dead days. Since the late 1950s the mists have begun
to fall. Legend's lustre is dimmed. While it may continue to figure in textbooks
and journalistic expositions, numerous intelligent critics now view Legend as
smug, uninformed, unhistorical, and analytically shallow. Some of these crit-
ics, the science bashers, regard the failure of science to live up to Legend's
advertising as reason enough to question the hegemony of science in contem-
porary society. I shall not be concerned with them, but with the critiques of
the Legend bashers, those who believe that Legend offered an unreal image
of a worthy enterprise.5
can go forward, we need an accurate picture of science, its goals, and its achievements.
This book can thus be seen as a prelude to a more wide-ranging study.
Legend's Legacy 1
in science were wrong, that the idealizations they employed were too simple,
or that their vocabulary needed enrichment. Instead, he suggested that the
entire project of finding a "calculus of scientific reasoning" ought to be
abandoned.
One way of developing that conclusion would be to claim that there are
no standards of good reasoning in science that are binding on all scientists at
all timesor at least that such standards would be insufficiently powerful to
endorse all the major decisions made in the history of science. In a celebrated
discussion, Paul Feyerabend (1970, 1975) maintained that Galileo's defense
of Copernicanism violated philosophers' favorite rules of good reasoning, and
he concluded that the only maxim that can be applied across all historical
instances is "Anything goes." Norwood Russell Hanson (1958) campaigned
for the theory-ladenness of observation, suggesting that there is no theory-
neutral body of evidence to which scientific theories must conform. So, by
emphasizing the lack of shared methodological rules and the inefficacy of
those few canons that are shared, by insisting on the variability of what is
taken as evidence, some critics of Legend made it seem appropriate to talk
of major scientific decisions as "conversion experiences."6 Because massive
underdetermination of belief by "objective" factors came to seem omnipres-
ent, there opened up a vacuum into which social explanations of scientific
behavior could be inserted. Instead of an ordered abode of reason, science
came to figure as the smoke-filled back rooms of political brokering. This was
not the way in which Kuhn intended to go, and, to the disappointment of
some who have drawn inspiration from him, he has continued to insist on a
set of commitments that scientists share, that cannot be articulated as rules
but that function in distinguishing good reasoning from bad.7
Critics of scientific rationality maintain not only that those who resisted
past decisions that we endorse as "correct" may have been just as reasonable
as their opponents but also that our endorsement itself is crucially dependent
on that initial decision. We should not comfort ourselves with the idea that
the choice among rival theories was temporarily underdetermined and that
those who prevailed wereluckilysubsequently vindicated by the exposure
of decisive evidence. Instead, we should recognize that either the choice
6. The phrase is Kuhn's (Kuhn 1962 151). In the later chapters of his book, Kuhn
assembles many of the principal arguments that others have deployed to reach far more
radical conclusions. He contends that observation is theory-laden (1962 chapter X) and
that shared methodological canons are too weak to determine scientific choices (1962
chapter XII). Yet the penultimate paragraph of the latter chapter suggests a reshaping of
traditional ideas about scientific rationality, not a decisive break. "Because scientists are
reasonable men, one or another argument will ultimately persuade many of them. But
there is no single argument that can or should persuade them all. Rather than a single
group conversion, what occurs is an increasing shift in the distribution of professional
allegiances" (1962 158). Chapters 6 and 8 later will develop some of the ideas at which
Kuhn seems to hint here.
7. See, in particular, chapter 13 of (Kuhn 1977). In Chapter 6, I shall develop a position
that has many affinities with Kuhn's attempt to steer a middle course between Legend and
relativism.
8 The Advancement of Science
actually made or its rival would have been self-sustaining: had matters gone
differently, our counterparts would have seen a history of success, congrat-
ulating themselves on the fact that subsequent evidence had supported the
initial decision. From this perspective, claims about the success and rationality
of science are a wonderful illusion, fostered by our ability to rewrite history
and to lose sight of the possibilities of discarded points of view. Legend's
praise of science is only the kind of boosterism that any proponent of any
"form of life" might engage in. Contemporary societiesat least those of the
industrialized worldhave made science part of their "forms of life," but
those are, in the end, no better or worse than the practices of the Azande,
the recommendations of homeopathy, or the promises of parapsychology.
They are simply ours.8
Despite the efforts of a few philosophers,9 little headway has been made
in finding a successor for Legend. If anything, recent work in the history of
science and in the sociology of science has offered even more sweeping ver-
sions of the original critiques. Philosophy of science, meanwhile, has enjoyed
a renaissance of studies of particular sciences.10 However, the contemporary
attention to a wide range of examples is typically eclectic: practitioners borrow
what general philosophical categories they need for the purposes of their
individual studies, attempting only to rely on concepts and distinctions that
seem to have survived Legend's demise.
Perhaps these individual studies (largely ignored by those who condemn
philosophy of science as defunct) represent the proper role for the discipline.
Perhaps there is no general picture of the sciences that philosophers can give.
Perhaps individual scientific disciplines and achievements are just thatin-
dividual, bound together by nothing more than "family resemblance."11 Al-
though these suggestions may offer sound advice, I am not yet ready to
abandon the search for generality. For without some substitute for Legend,
minimal, eclectic, and tacit though it may be, it is not clear that individual
studies of particular sciences can be pursued. Moreover, the arguments of
the Legend bashers deserve answers or acknowledgment of their correctness.
The goal of this book is to draw a picture of how science advances, using
8. For a strong version of this type of relativism, see (Barnes and Bloor 1982) and
(Collins 1985). The case of parapsychology is discussed in some detail in (Collins and Pinch
1982). Similar ideas ares advanced in (Feyerabend 1975,1979,1987). The notion of a "form
of life," now much in vogue among some sociologists of science, descends from (Wittgen-
stein 1953). A locus classicus for the application of this notion to the history of science and
for many of the themes noted in the text is (Shapin and Schaffer 1985).
9. Most notably Imre Lakatos (1970), Larry Laudan (1977, 1984, 1990), and Dudley
Shapere (1984).
10. Not only in philosophy of physics, to which Legend's main elaborators made major
contributions, but increasingly in philosophy of psychology and philosophy of biology as
well.
11. (Wittgenstein 1953 67). One might abandon the attempt to offer a general phil-
osophical vision of science for Wittgensteinian reasons, holding that there is no essence to
the enterprise, while commending the activity of studying and characterizing particular
scientific contributions or debates. Arthur Fine has sometimes advocated this kind of view
in lectures.
Legend's Legacy 9
the commonsensical ideas that underlie Legend, the insights of Legend's
critics, as well as the contributions of contemporary philosophers, historians,
sociologists, and cognitive scientists. The chief questions I shall address are
those that Legend took as central: What is scientific progress? How is science
pursued rationally? Problems about progress occupy Chapters 4 and 5. The
goal of Chapter 4 is to provide a definition of scientific progress, or, more
exactly, to characterize some varieties of scientific progress and to connect
them to aims that I take to be worth satisfying. Chapter 5 attempts to defend
my characterizations against criticisms to the effect that the aims I specify are
incoherent or unattainable, and I try to rebut charges that science does not
make progress in the senses I pick out. In Chapter 6 I turn to questions about
scientific rationality, attempting first to specify what an idealor idealsof
rationality ought to be. The traditional project of articulating canons of ra-
tionality for the epistemic projects of individuals is taken up in Chapter 7.
Finally, in Chapter 8, I turn to the important, though largely neglected, issue
of how various combinations of individual epistemic strategies might advance
or retard the community's epistemic enterprise.12
Before these accounts can be developed, some preliminary work is nec-
essary. My attempts to combine what I take to be important (and currently
unappreciated) insights of logical empiricism with the apercus of historians
and sociologists will rest on a novel way of idealizing the phenomena. Instead
of thinking of science as a sequence of theories and of theories as sets of
statements, I shall offer a multi-faceted description of the state of a science
at a time. Moreover, I shall be concerned to treat the growth of science as a
process in which cognitively limited biological entities combine their efforts
in a social context. Placing the knowing subject firmly back into the discussion
of epistemological problems seems to me to be the hallmark of naturalistic
epistemology.13 Chapter 3 is my attempt to construct a framework for natu-
ralism in the philosophy of science.
The constructions of Chapter 3 will be guided by a review of some phe-
nomena. In Chapter 2 I shall offer an extended illustrative example, on which
subsequent discussions will be able to draw.14 I shall examine some facets of
12. The idea of understanding how social institutions might promote knowledge has
been largely ignored in traditional epistemology. Pans of the problem begin to emerge in
(Kuhn 1962), in (Sarkar 1982), and in (Hardwig 1985). Some suggestive points are made
in (Fuller 1988), and Alvin Goldman (1987) offers a general overview of the issues. Some
highly pertinent source material with interesting interpretations can be found in (Ghiselin
1989) and (Hull 1988).
13. This is a common feature of the enterprises of (Goldman 1986) and (Quine 1970a),
although they apparently disagree on the possibility of undertaking normative inquiries in
epistemology. The present book exemplifies my kinship on this issue with Goldman (see
also my 1983). Other accounts of what makes for a naturalistic epistemology can be found
in the literature. Laudan (for example) conceives of naturalistic epistemology as committed
to the notion that methodological rules are empirical (see Laudan 1989). General discussions
of the character of naturalistic epislemology can be found in (Kornblith 1985), (Giere 1985),
(Kim 1988), and (Kitcher 1992).
14. Other examples from past and present science will be discussed more briefly in later
chapters, but it seems to me to be impossible to elaborate a convincing picture without
providing one, fairly detailed, illustration. Ultimately, philosophical pictures should be
10 The Advancement of Science
the history of evolutionary ideas, from the midnineteenth century to the
present, with a view to fixing ideas about goals, methods, progress, rationality,
individual scientific behavior, and the social structure of science.
All the discussions that follow are suffused with a vision of philosophy of
science that should be made explicit. In my judgment, philosophical reflections
about science stand in relation to the complex practice of science much as
economic theory does to the complicated and messy world of transactions of
work, money, and goods. Much traditional philosophy of science, in the style
of some economic modeling, neglects grubby details and ascends to heights
of abstraction at which considerable precision and elegance can be achieved.
We should value the precision and elegance, for its own sake, for its estab-
lishing a standard against which other efforts can be judged, and for the
possibility that extreme idealizations may lay bare large and important fea-
tures of the phenomena. But, like ventures in microeconomics, formal phi-
losophy of science inevitably attracts the criticism that it is entirely unrealistic,
an aesthetically pleasing irrelevancy. To rebut such chargesor to concede
them and to do better service to philosophy's legitimate normative project
we need to idealize the phenomena but to include in our treatment the features
that critics emphasize.
Legend's legacy is the task of recognizing the general features of the
scientific enterprise, most importantly by scrutinizing the apparent progres-
siveness of science, the seeming individual rationality of individual scientists,
and the collective rationality of the scientific community. In light of the serious
challenges to Legend, there is no doubt that the task is difficult. However,
recurrent pronouncements of the death of philosophy notwithstanding, it is
too important to be discarded.
tested against a range of cases, historical and contemporary. Articulating the methodology
underlying this test procedure is itself a formidable taskand, moreover, one that cannot
be detached from our philosophical picture of science.
2
Darwin's Achievement
11
12 The Advancement of Science
to a quartet: Huxley, Hooker, and Lyell had been joined by Asa Gray. Darwin concluded
a letter to Gray, "It is the highest possible gratification to me to think that you have found
my book worth reading and reflection; for you and three others I put down in my own
mind as the judges whose opinions I should value most" (F. Darwin 1888 II 272). There
is no inconsistency here, but the juxtaposition of the passages reveals Darwin's tact.
4. For Darwin's securing of a place in the British scientific establishment, see (Moore
1985 450-452, 460-462) and (Manier 1978). The mixed intellectual and personal character
of Darwin's relations with his colleagues is beautifully captured in Hooker's account of
visits to Down (see F. Darwin 1888 I 387-388, Ospovat 1981 91-94).
5. Huxley offers a clear account of how he came to espouse Darwin's theorysee his
(1896 246). He also attributes the rapidity of the Darwinian revolution to the arguments
of the Origin (1896 286). Of course, Huxley's stress on reason and argument may be self-
deception. It is interesting to note that, in the 1850s, before he became a personal friend
of Darwin's, Huxley offered a ranking of the leading British biologists of the day. Darwin
figures in the second tier: he is described as "one who might be anything if he had good
health" (L. Huxley 1913 I 94).
6. The most explicit formulations of this type of critique are given in (Ho and Saunders
1984) and in several essays in (Ho and Fox 1988). (See, in particular, the introduction and
the article by Brian Goodwin in the latter work.) But there are less strident attacks on
neo-Darwinism that question cumulativist views of the history of evolutionary biology:
examples are (Gould 1982, Eldredge 1985, Levins and Lewontin 1985, Stanley 1984).
Darwin's Achievement 13
the point being made. My guiding analogy is with a sketch that lays down
the main lines around which the detailed picture is subsequently drawn. Phil-
osophically oriented history should not "reconstruct" in the sense of drawing
lines that would have to be altered in a more detailed presentation.7
2. Before Darwin
To understand the context in which Darwin's ideas were received and debated,
we need to identify some major themes in early nineteenth century thinking
about the history of earth and of life. From the close of the eighteenth century
on, it had become increasingly evident to the cognoscenti that the time span
for life on earth was far longer than had traditionally been assumed. In the
early decades of the nineteenth century Georges Cuvier established the phe-
nomenon of extinction, and, in the wake of his painstaking paleontological
work, natural historians formulated an increasingly detailed picture of a suc-
cession of organisms.8 But the problem of identifying the process(es) driving
the succession of organismsDarwin's "mystery of mysteries" (1859 1)was
bypassed in favor of other biological projects.
The tradition of natural theology, especially strong in Britain,9 was ded-
icated to the idea of exhibiting the exquisite fit of animal to environment.
The utilitarian conception of life, favored by the British who conceived of
the Creator as engineer extraordinaire, contrasts with the approach of the
German Naturphilosophen. Influenced by the ideas of the poet Johann Wolf-
gang Goethe, the Romantic morphologists sought archetypes (Urbilde) for
the major groups of organisms. The principal task of comparative biology,
as they conceived it, was not to discern the intricate ways in which organic
structures contributed to the well-being of their bearers but to expose the
underlying communities of structure. Ultimately, all organisms were to be
organized by considering the ways in which each diverged from the corre-
sponding archetype, so that the plan of creation would be exhibited by un-
AnalogueA part or organ in one animal which has the same function
as a part or organ in a different animal.
HomologueThe same organ in different animals under every variety
of form and function (1848 7).
What is most significant about this double introduction of terms is the implicit
recognition of two types of affinity in animal structures. Some similarities
(those on which Cuvier insisted) represent commonality of function. Others
(those Geoffroy tried to discern everywhere) indicate community of type.
As his ideas developed in the 1830s and early 1840s, Owen revived the
concept of the archetype, regarding it as a formal distillation of the homologies
revealed in his specimens. The task of the comparative anatomist was to
include more than the simple tracing of individual homologies. Ultimately,
10. In the works of Lorenz Oken and Karl von Baer, this approach was elaborated
with considerable imagination and some illumination. Embryological similarities were in-
terpreted as signs of sameness of archetype, and, indeed, the process of development could
be construed in terms of divergences from archetypal form. As Ernst Mayr points out,
Oken's fanciful ideas were frequently erroneous, although his approach "was actually very
productive in arthropod morphology, helping to homologize mouth parts and other cephalic
appendages with extremities" (1982 458). A sympathetic and insightful account of the ideas
of the Naturphilosophen is provided by Gould in his (1977).
11. Geoffroy advocated a "philosophical anatomy," an approach to comparative mor-
phology that had obvious affinities with the ideas of the Naturphilosophen but that included
important contributions of his own. In particular, he introduced a version of the homology/
analogy distinction.
Darwin's Achievement 15
Owen hoped to display the archetypesstructural plans or Platonic Ideas
that underlie the diversity of organisms. Functional biology was to investigate
the ways in which the plan is modified in its embodiment, so that organisms
are adapted to their environments.
Owen's method of fusing the ideas of some of his predecessors could be
extended to a view of the history of life that articulated the tradition of natural
theology in a novel way. Paleontology reveals a succession of organisms, each
of which embodies one of a limited stock of archetypes. To uncover the plan
of Creation is to trace the ways in which the sequence of living things through
time articulates the timeless Ideas, how each organism is exquisitely adapted
to its environment and yet how each is fundamentally structured on one of
a small number of plans. To appreciate both the economy of archetypes and
the harmonies of adaptation is to recognize the beneficence and wisdom of
the Creator.12
In the early 1830s Charles Lyell published his magisterial Principles of
Geology (which Darwin was to read aboard the Beagle), ordering the fossils
of the Cenozoic and offering, contra Cuvier, a picture of the history of the
earth in which changes in environment result from the action of presently
observable forces over immense periods of time. Within the next fifteen years,
several outstanding puzzles of the fossil record were resolved, and the pa-
leontologists of the 1840s and 1850s could, with some confidence, pronounce
on the relative ordering of strata, assigning fossils to their temporal ranges.
Commitment to the idea of extinction and renewal of the earth's fauna en-
tailed, of course, commitment to the existence of some process through which
new species were produced. There were, apparently, two options. Either the
new species were descended from those now extinct (the transmutationist or
evolutionary hypothesis) or they had been called forth in a new act of creation.
The sophisticated drew back from speculation. Lyell declined to pronounce
on the processes by means of which novel species have appeared. Similarly,
Owen maintained a studied agnosticism. It was enough for both of them to
declare that the types of organisms that emerge are determined by "secondary
causes."
From a post-Darwinian point of view this failure to attempt a resolution
of the issue appears strange. Surely one must ask how new species have
originated! To dissolve the mystery, we need to note two aspects of the pre-
Darwinian situation. First, with transmutationism in disrepute, partly as a
result of Lamarck's formulation of it, partly because of the force of the familiar
objections that Cuvier had marshaled so effectively, inquiry into the pro-
cess(es?) of species origination threatened to degenerate quickly into irre-
sponsible speculation. Second, both Lyell and Owen could pose for themselves
and their colleagues manageable and apparently significant research prob-
lemsthe replacement of appeals to catastrophes with "uniformitarian" ex-
12. See (Owen 1849). Both Rudwick and Michael Ruse (1979) offer clear accounts of
Owen's ideas and testify to their power and subtlety. See also (Russell 1916).
16 The Advancement of Science
planations, the continued description of the fossil record, the display of hom-
ologies and, ultimately, the delineation of the unfolding of the archetypes in
the history of life.13
But, even before Darwin, there were some areas within biology in which
considerations of pattern seemed insufficient, in which considerations of his-
tory became evident, and in which speculations about creation surfaced. Just
as the early nineteenth century witnessed a great increase in understanding
of the distributions of animals (and plants) through time, so too the voyages
of exploration (and the reports of the naturalists who sailed on them) yielded
a greatly enriched stock of biogeographical phenomena. The discoveries of
very different faunas on different continentsespecially the Australian mar-
supialsmade it very difficult to maintain that there had been a single center
of creation from which (by some mysterious principle of distribution) the
animals had dispersed. For early nineteenth century biogeography, the prin-
cipal questions were to understand what the spatial distributions of plants
and animals actually are and why they are that way. In its latter project,
biogeography became historical and the history involved facing the topic of
the origins of species.
The leading biogeographers of the early nineteenth century invoked the
notion of "centers of creation," a concept that had been proposed in the
eighteenth century and had quietly been accepted as a way of modifying the
Biblical story of creation.14 Thinking of species as immutable, pre-Darwinian
biogeographers were severely constrained in their historical analyses. Similar
contemporary species have similar, but distinct, populations of founders, and,
after one has reconstructed their histories of dispersal, one is forced to con-
clude that they were separately created, perhaps at the same center. What
principles underlie the origination of some groups of similar species in certain
places, and the origination of different groups of similar species in different
places, even when the centers of creation have, apparently, quite similar
13. Similarly, in both the French and German research traditions, one can find the
posing of questions about pattern in the history of life that are independent of problems
about the process(es) through which new species have emerged. See, for example, the
influential essay on classification by Henri Milne-Edwards (1844).
14. For a concise and informative discussion of this concept and its employment in pre-
nineteenth century biogeography, see (Mayr 1982 439-443). Alphonse de Candolle, whose
discussion of the geographical distribution of plants was recognized by Darwin for its
comprehensiveness, proposed that there are more than twenty separate botanical regions.
Similarly, Edward Forbes offered an explanation of the distributions of British plants and
animals by positing separate centers of creation. Both Forbes and de Candolle had begun
to naturalize biogeography by seeking historical accounts of current distributions. Both
rejected, with good reason, the once popular idea that the distributions of species could
be understood teleologically by noting that each species range covered just those environ-
ments to which the species was best suited. By the 1840s and 1850s it was clear from
abundant examples that species sometimes were absent from regions in which they might
thrive and present in places that were less than ideal for them. De Candolle and Forbes
(and Lyell before them) saw the present distributions of plants and animals as resulting
from processes of dispersal. Yet, after the history of dispersal had been retraced, the
biogeographer inevitably came to the point of origin, the center of creation.
Darwin's Achievement 17
15. Lyell was convinced that the answer depended on unnoticed features of the eco-
logical and geological conditions, that there are laws relating the characteristics of the
species produced to the conditions of their places of origin. Alphonse de Candolle, by
contrast, maintained that appeal to climatic and physical conditions alone would prove
insufficient. See (De Candolle 1855 33-34, translated in Ospovat 1981 16).
16. For discussions of Lamarck's evolutionary views and the differences with Darwin's
ideas, see (Hodge 1971, Burkhardt 1977, Ruse 1979, and Mayr 1982 343-359). Darwin's
reactions to Lamarck are atypically harsh. See (F. Darwin 1888 II 23, 29, 39, 215).
17. Mayr (1982 382) points out that Chambers' writing outsold in the first decade the
classic works of Lyell and of Darwin in their (respective) first decades.
18. For Baden Powell's judicious claims, many of which point towards the arguments
of the Origin, see his (1855 360, 372-733, 376, 379, 387, 401, 413-427). These passages
also make clear the influence of Owen's agnosticism about transmutation, which often leads
Baden Powell to draw back from conclusions that Darwin would embrace.
18 The Advancement of Science
Baden Powell's cautious defense of the possibility of transmutation made it
apparent that Owen's ideas might be extended to an evolutionary account of
the history of life. By 1859, Darwin had been at work on such an account for
over twenty years.
3. Darwin's Innovations
Virtually everyone would agree that the Origin offers something new, a new
theorythe theory of evolution by natural selectionor several new theo-
ries.19 Attempts to specify what the novel theory is are inevitably influenced
by general ideas about scientific theories. A battered and truncated version
of the "received view" (scientific theories are axiomatic deductive systems)
lingers on. Even those who are skeptical about the need for distinctive the-
oretical vocabulary, or correspondence rules, or axiomatizability, or any of
the other ingredients of the logical empiricist picture,20 are likely to suppose
that any scientific theory worthy of the name must consist of a set of state-
ments, among which are some general laws (laws that set out the most
fundamental regularities in the domain of natural phenomena under inves-
tigation), and that such laws should be used to derive previously unaccepted
statements whose truth values are subject to empirical determination.21 When
this residual thesis is applied to the case of Darwin, we are led to expect that
the Origin advances some collection of new general principles about orga-
nisms. After all, what else could Darwin's theory be?22
19. Ernst Mayr (1982,1991) distinguishes five theories that are advanced in the Origin.
While it is important to separate Darwin's basic commitment to evolution (the thesis that
species are connected by descent with modification) from his views about the mechanisms
of evolutionary change (most prominently his claims on behalf of natural selection), it
seems to me that Mayr multiplies theories beyond necessity.
20. For a sustained account of the "received view" and its problems, see Fred Suppe's
comprehensive introduction to his (1977).
21. This minimal conception of scientific theories is often implicit in the historical
literature, where it encourages the effort to find some short formulation of Darwin's new
theory. See, for example (Mayr 1982, Ospovat 1981, Ghiselin 1969, Ruse 1979, and many
of the essays in Kohn 1985). For many historical purposes little harm comes of this practice.
But if one wants to be precise about the differences between Darwin and his predecessors
and contemporaries and if one hopes to understand the reasoning of the Origin, then, as
we shall see, the minimal conception poses severe difficulties.
The minimal conception can also be found in the practice of many philosophers who
offer accounts of scientific changeit is, for example, espoused by Lakatos, Laudan, and
Shapere. However, both Laudan and Shapere break with the popular supposition that the
state of a science at a time can be encapsulated in the set of accepted theories. This is a
salutary move and leads to positions with some kinship to that which I shall develop.
22. One answer, offered by Elisabeth Lloyd in her (1983), is that the theory is a
collection of models (or, perhaps, model types). I had originally thought that Lloyd's
advocacy of the "semantic conception" of scientific theories (a conception advanced and
defended by Suppe (1972), Suppes (1965), Sneed (1971), and van Fraassen (1980) among
others) was at odds with my own reconstruction. After numerous discussions with her, I
am no longer sure about this. I have not found it helpful to use the standard terminology
Darwin's Achievement 19
From these principlesmore exactly from (2), (3), and (4)one can obtain
by a plausible argument
(5) Typically, the history of a species will show the modification of that
species in the direction of those characteristics which better dispose their
bearers to survive and reproduce; properties which dispose their bearers
to survive and reproduce are likely to become more prevalent in suc-
cessive generations of the species (The Principle of Natural Selection;
Origin chapter 4).
that most stand in need of defense and confirmation, and that the arguments
assembled by the innovative theorist should be directed at the fundamental
principles of the novel theory.
Virtually all of Darwin's opponents would have accepted (l)-(4). None
of the great naturalists of the midnineteenth century would have denied
and none should have deniedthat species vary, that the increase of a species
is checked, that some variation affects characters relevant to the ability to
survive and reproduce, that many properties are heritable. Moreover, they
saw the force of the argument for (5), assenting to the idea that natural
selection has the power to adjust the distribution of traits within a species,
eliminating variants whose characteristics render them less able to compete
in a struggle for limited resources. What was in dispute was not so much the
truth of (l)-(5) as their significance.23 For the committed Darwinian, these
principles were the key to understanding a vast range of biological phenom-
ena, and the principal theoretical and argumentative work of the Origin
consists in showing how the seemingly banal observations about variation,
competition, and inheritance might answer questions that had previously
seemed to be either beyond the scope of scientific treatment or else only
addressed in a partial and limited way by Darwin's predecessors.
Acceptance of (l)-(5) is compatible with the doctrine of fixity of species
(which states that species are closed under reproduction). However, Darwin
did not simply accept (l)-(5) and add the historical claim that lineages (ances-
tor-descendant sequences of organisms) have been modified to the extent that
they embrace members of different species. Owen seems to have defended
something like that view.24 Nor did Darwin simply conjoin (l)-(4) with the
denial of the fixity of species and the vague declaration that natural selection
has been the primary force of evolution. The Origin contains a novel and
well-articulated theory precisely because it fuses (1)(5) with the suggestion
that species are mutable to fashion powerful techniques of biological expla-
nation. If we are to make plain the nature of Darwin's innovation, we need
a new way of thinking about the complex entity that is the state of a science
at a time. Reconstructing the Origin will provide an exemplar.
I shall start from the idea that the Origin proposes an explanatory device,
aimed at answering some general families of questions, questions which Dar-
win made central to biology, by presenting and applying what I shall call
Darwinian histories. To fix ideas, I shall characterize a Darwinian history in
a preliminary way as a narrative which traces the successive modifications of
a lineage of organisms from generation to generation in terms of various
23. I am grateful to Malcolm Kottler for pointing out to me that there is a parallel with
the debates about Lyell's geological views, in that what was primarily in dispute was the
significance of Lyell's claims about the time scale. See (Rudwick 1970).
24. Owen's complicated views about evolution and natural selection can be recon-
structed from his anonymous review of the Origin (which frequently compares Darwin
unfavorably with "Professor Owen"). See (Hull 1973b 175-213, especially 196, 201). I shall
try to make clear in the text following what the differences between Darwin and Owen
were and why Owen was so bitterly critical of Darwin's proposals.
Darwin's Achievement 21
factors, most notably that of natural selection. The main claim of the Origin
is that we can understand numerous biological phenomena in terms of the
Darwinian histories of the organisms involved.
Consider first issues of biogeographical distribution. For any group G of
organismsperhaps a species, a genus, or some higher taxonwe can identify
the range of that group. With respect to any such group we can envisage a
complete description of its history. From the Darwinian perspective, this
historical description will trace the modification of the current group from its
ancestors, revealing how properties change along a (possibly branching) lin-
eage, and how, as these changes occur, the area occupied by members of the
group alters. Darwinian histories provide the bases for answers to biogeo-
graphical questions.
Darwinian histories provide the bases for acts of explanation, and, con-
fronted with a practical question of biogeographical distribution, incomplete
knowledge of a Darwinian history will enable us to offer an answer. When
we ask why a group G occupies a range R, we typically have a more particular
puzzle in mind. For example, someone who wonders why so many marsupial
species are found in Australia is likely to be perplexed by the fact that so few
are found elsewhere. Perplexity is relieved by outlining the Darwinian history
of the marsupialsdescribing how they were able to reach Australia before
the evolution of successful placental competitors, how the placentals were
able to invade many marsupial strongholds, and how the placentals were
prevented from reaching Australia. Here and in other instances,25 a general,
unfocused, explanation-seeking question is determined in context as a more
precise request for information. The request can be honored by abstracting
from the Darwinian history, so that the needed information can be given,
despite considerable ignorance about the details.
It is helpful to contrast the biogeographical proposals of the Origin with
the approaches reviewed in the last section. One (creationist) possibility would
resist the appeal to history altogether in favor of teleological answers to
questions about distribution: because each group of organisms was created
to inhabit a particular region, and because it has always inhabited that region,
our understanding of the phenomena of distribution is advanced by recog-
nizing those features of the organisms in the group that fit them to live where
they do. As I mentioned previously, by 1859 this approach had fallen into
well-deserved disrepute. It was well known that organisms transported by
humans could thrive in areas that they had previously been unable to reach,
and naturalists knew of other cases in which organisms seem ill-suited to their
natural habitat (a popular example was that of those woodpeckers who inhabit
treeless terrain).
More promising was the historical approach developed in most detail by
25. For example, the question why the birds known as "Darwin's finches" are confined
to the Galapagos, a question that has played an exemplary role in the history of Darwinian
biogeography. Ironically, as Frank Sulloway's brilliant detective work has shown (Sulloway
1982), it now appears that Darwin left the Galapagos without any appreciation of the
significance of the famous finches.
22 The Advancement of Science
de Candolle and by Forbes. On this view, we answer questions about bio-
geographical distribution by seeing the current range of an organismic group
as the result of a process in which an unmodified (or relatively unmodified)
sequence of organisms has dispersed from an original "center of creation."
The trouble is that this approach only postpones perplexity, for we need some
scheme for explaining the distribution of original centers of creationor at
least some way of making the ways of the Creator seem less capricious. Darwin
makes the point forcefully:
But if the same species can be produced at two separate points, why do we
not find a single mammal common to Europe and Australia or South Amer-
ica? The conditions of life are nearly the same, so that a multitude of Eu-
ropean plants and animals have become naturalised in America and
Australia; and some of the aboriginal plants are identically the same at these
distant points of the northern and southern hemispheres? (Origin 352-353;
see also Origin 394)
Darwin's challenge is to provide a comprehensible distribution of centers of
creation that will allow for the disconnected distribution of the plants common
to the hemispheres, while explaining the failure of the mammals to radiate
into regions for which they are well suited. The thrust is that creationists will
ultimately be forced into conceding that the distribution of original centers
of creation is inexplicable. By contrast, as Darwin emphasizes, the theory of
evolution claims for scientific explanation questions which rival theories dis-
miss as unanswerable.
I have begun with the example of biogeography, because it is the case on
which Darwin often lays the greatest stress, suggesting that it was reflection
on biogeography that originally led him to the theory of evolution.26 Bio-
geography is only part of the story. The rest is more of the same kind of
thing. With respect to comparative anatomy, embryology, and adaptation,
Darwin also provided strategies for answering major families of questions.
Consider comparative anatomy. Here one major task is to provide answers
to questions of the following general form:
26. (F. Darwin 1888 I 336, II 34; F. Darwin 1903 I 118-119). Darwin's elder brother,
Erasmus, confessed, "To me the geographical distribution, I mean the relation of islands
to continents is the most convincing of the proofs, and the relation of the oldest forms to
the existing species" (F. Darwin 1888 II 233). Huxley describes the importance of bio-
geography in his own reception of the theory of evolution (1896 276). Darwin's public
account of the role of biogeography in his own thinking occurs in the opening sentences
of the Origin.
Darwin's Achievement 23
by relating the histories of descent from a common ancestor (in which P was
also present). In cases where P is a mere analogy, its common presence will
be understood by tracing the history of the emergence of P along the lineages
leading to G and G', perhaps showing how this was the result of similar
environmental pressures.27 Darwin reviewed the classic cases. Similarities in
the bone structure of the forelimbs in various mammalian groupsmoles,
seals, bats, ruminantsare to be understood in terms of descent from a
common ancestor. By contrast, the existence of wings in birds, bats, flying
reptiles, and insects, is understood by recognizing the paths which these groups
have followed in evolving the ability to fly.28
Plainly there are important similarities between the programs for com-
parative anatomy envisaged by Darwin and by Owen. Both employ a dis-
tinction between homology and analogy and propose different patterns of
explanation for homologous and analogous traits. But where Owen invokes
the presence or absence of a shared archetype and remains carefully non-
committal on the historical processes that have given rise to the groups under
comparison, Darwin appeals to a shared ancestor and outlines a historical
explanation. As in the case of biogeography, while relating the complete
Darwinian histories of the groups involved would provide an ideal answer to
a question about the relationships among them, our practical questions about
the similarities among organisms do not require such detail. Quite frequently,
the question of why two groups of organisms agree in a morphological prop-
erty stems from puzzlement that organisms so different in other respects
should share the morphological trait in question. If the property is a homology,
the puzzle is resolved by outlining enough of the Darwinian histories of the
organisms to reveal the main lines of their modifications from a common
ancestor. Similarly, in the case of analogy, we need to tell enough of the
Darwinian histories to understand how a similar feature has been produced
in unrelated lineages.
As with biogeography, Darwin could easily contrast his approach to com-
parative anatomy with the available rivals. By pointing out such differences
as those between the blind cave insects of Europe and America (Origin 138),
he could pose a deep challenge to the sophisticated program for comparative
anatomy advanced by Owen. On Owen's conception, we would explain the
differences between North American and European cave insects by suggesting
that the former realize certain "north american" archetypes while the latter
27. Note that at this point the explanation goes beyond simple recognition of ancestor-
descendant relationships. Darwinian histories that trace the emergence of analogous traits
by appealing to common environmental pressures must venture claims about the causes of
evolutionary changesperhaps by appealing to natural selection, perhaps by invoking one
of Darwin's other mechanisms (e.g., direct induction by the environment).
28. Darwin provides a very clear account of the homology/analogy distinction and is
well aware that his theory enables him to refine the concept of homology; see Origin 427
and (F. Darwin 1903 I 306). Note that, even though Owen's version of the distinction was
more sophisticated than those of his predecessors, the analysis he offered still suffered from
vagueness.
24 The Advancement of Science
embody distinct "european" archetypes. But now there is a puzzle about why
different archetypes are realized in the caverns of North America and in the
caverns of Europe. Owen is committed to the idea that there are "secondary
causes" that govern the emergence of the distinct species on the two conti-
nents, but he remains agnostic about whether the process of emergence in-
volves reproductive continuity or creation de novo. However, the two options
envisaged are not on explanatory par. If the cave insect species have evolved
from ancestral species of insects, then we can understand the structural re-
lations. But if they stem from separate processes of special creation, then the
persistence of distinct "north american" and "european" archetypes in the
two regions is quite mysteriousthere is nothing in the environmental con-
ditions to call forth different archetypes, and we seem to be forced to some
odd hypothesis of parallelism.
The third example I shall consider is historically crucial, in that it repre-
sents the most promising field for the tradition of natural theology. Darwin
confesses that his theory could not be admitted as satisfactory "until it could
be shown how the innumerable species inhabiting this world have been mod-
ified, so as to acquire that perfection of structure and coadaptation which
most justly excites our admiration."29 However, he proposes that questions
of adaptation, like questions of biogeography and comparative morphology,
can be answered by rehearsing the historical process through which the ad-
aptation emerged. The general form of question to be addressed is
29. Origin 3. Richard Lewontin (1978) perceptively discusses the way in which prior
emphasis on the problem of design made the discussion of adaptation central to Darwin's
evolutionary thinking, and how this fact, in turn, led Darwin to stress the selectionist
commitments of his theory.
30. Here I deliberately overdraw the adaptationist commitments of Darwin's theory. I
shall consider below whether Darwin allows a more pluralistic approach to the evolutionary
explanation of apparently beneficial characteristics.
Darwin's Achievement 25
We need not marvel at the sting of the bee causing the bee's own death; at
drones being produced in such large numbers for one single act and then
being slaughtered by their sterile sisters; at the astonishing waste of pollen
by our fir-trees; at the instinctive hatred of the queen bee for her own fertile
daughters; at ichneumonidae feeding within the live bodies of caterpillars;
. . . (Origin 472)
Other passages descant on the "Panda's thumb theme,"31 the existence of
many cases in which it is evident that natural contrivances fall far short of
the standards of good design we would expect from a competent engineer,
and in which it is more plausible to suppose that the available materials
dictated a clumsy solution to a design problem.
He who believes in the struggle for existence and in the principle of natural
selection, will believe that every organic being is constantly endeavouring to
increase in numbers; and that if any one being vary ever so little, either in
habits or structure, and thus gain an advantage over some inhabitant of the
country, it will seize on the place of that inhabitant, however different it may
be from its own place. Hence it will cause him no surprise that there should
be geese and frigate-birds with webbed feet, either living on the dry land or
most rarely alighting on the water; that there should be long-toed corncrakes
living in meadows instead of in swamps; that there should be woodpeckers
where not a tree grows; that there should be diving thrushes, and petrels
with the habits of auks. (Origin 186)
This theme receives its most detailed treatment in Darwin's book on orchids
characterized as "a 'flank movement' on the enemy."32 Again, it points toward
the same moral: questions that rival approaches must dismiss as unanswerable
can be tackled by adopting the Darwinian perspective.
4. Darwin's Patterns
I have claimed that the Origin offers a class of problem-solving patterns,
aimed at answering families of questions about organisms by describing the
histories of those organisms. The complete histories will always take a par-
ticular form in that they will trace the modifications of lineages in response
to various factors"Natural Selection has been the main but not exclusive
means of modification."33 I now want to give a more precise account of the
notion of a Darwinian history.
31. See the title essay in (Gould 1980). I think that Gould is correct to view Darwin's
sounding of this theme as central to his case for evolution.
32. See (Darwin 1862). Ghiselin provides a penetrating analysis of the argumentative
strategy I have attributed to Darwin and of its significance in Darwin's defense of evolution.
See his (1969 137). Darwin's military characterization of the role of the orchid book is
given in (F. Darwin 1903 I 202).
33. Origin 6. The difficulty of interpreting Darwin's sentence is obvious. Natural se-
lection might be heralded as the force that produces most evolutionary changes, or, perhaps,
as the force that produces the most important evolutionary changes. An alternative con-
26 The Advancement of Science
Darwinian histories are structured texts. I shall try to exhibit the structures
that Darwin's paradigm explanations exemplify. Let us start with a notion of
Darwinian history that is minimal in the sense of embodying the fewest as-
sumptions about the tempo and mode of evolutionary change. This conception
can be characterized as follows:
This minimal conception allows for evolutionary change, for the property that
frequencies may vary from generation to generationindeed, traits initially
absent may ultimately be found in every member of the groupbut it does
not offer any account of why this change occurs.
Consider, for example, some questions about biogeographical distribution.
To understand why the Galapagos contains endemic species of finches which
are similar to mainland South American forms it is sufficient to know the
history of descent with modification without speculating about the modifying
factors.34 The structure of this explanation is as follows: we build on a minimal
Darwinian history by using claims about the dispersal powers of the organisms
and their current ranges to derive descriptions of the ranges in the next
generation (or, more exactly, to derive statements that give the probabilities
of ranges). Behind the explanation is a schematic form, instantiated in many
Darwinian answers to biogeographical questions, in which we trace sequen-
tially the changing geographical distributions of properties (including those
properties that are crucial to species differences) and derive, at each stage,
the current range of organisms with particular complexes of traits from as-
criptions of powers of dispersal and the ranges in the previous generation.
Similarly, some questions about the relationships among groups of or-
ganisms can be addressed by rehearsing histories of descent with modification
that do not explore the causes of the changes that have occurred along the
lineages involved. Darwinian explanations of the presence of homologies take
a very simple form, which we can exhibit as follows:
ception would be to suppose that the modifications produced through other causal processes
are somehow impermanent, so that the large-scale course of evolution follows the trajectory
laid down by selection. Finally, one might concentrate on those evolutionary changes that
produce new taxa (for example, speciation events), understanding these as being effected
by natural selection.
34. I do not claim that this will be possible with respect to all questions about biogeo-
graphical distribution. There are obviously many instances in which understanding the range
of a species will involve recognition of the competitive relations with other species, and in
which resolution of the biogeographical questions will turn on issues of coadaptation.
Nonetheless, it is sometimes possible to answer such questions without investigating issues
of adaptation, and in such cases the history of descent with modification will suffice.
Darwin's Achievement 27
COMMON DESCENT
Question: Why do the members of G, G' share P?
Answer:
(1) G, G' are descended from a common ancestor G0.
(2) G0 members had P.
(3) P is heritable.
(4) No factors intervened to modify P along the G0-G, G0-G'
sequences.
Therefore (5) Members of G and G' have P.
I suggest that this elementary pattern underlies the simplest illustrations that
Darwin provides of natural selection: the explanations of the swiftness and
slimness of wolves, of the presence of two varieties of wolf in the Catskills,
the excretion of "sweet juices" by some plants, and many others.39
The intuitive idea behind SIMPLE INDIVIDUAL SELECTION is that one explains
the presence of some traits (adaptations) by beginning with their initial ap-
pearance in some ancestral group, identifying the factors that favored them,
and tracing their increasing prevalence. In many instances, especially when
he is concerned with complex characteristics, Darwin's explanations have a
38. Obviously the elementary pattern that I have suggested could be refined in various
ways that would make it more adequate as a representation of the simplest accounts of
selection that Darwin gives in the Origin. In particular, it is evident that the notion of
"sufficiently many" generations is tied to the amount by which the expected reproductive
success of P bearers exceeds that of the competition (the fitness difference, as we would
put it). Although Darwin was evidently aware of this point, his analyses in the Origin are
nonquantitative, so that the use of the vague term "sufficiently many" is, in some ways,
closer to his actual explanations.
39. The cases I have mentioned are presented at Origin 90-92. There are numerous
others throughout the Origin.
Darwin's Achievement 29
more intricate structure. Instead of simply showing how the final version of
a traitperhaps an anatomical structure like an eye or a wingbecame prev-
alent, the task is to understand how ancestral forms were successively modified
in the direction of the current characteristic. We can think of this as a pattern
that stitches together a number of instantiations of SIMPLE INDIVIDUAL
SELECTION
The selectionist patterns I have discussed take variation for granted. There
is no systematic attempt to explain the presence of the variant traits in G0 by
deriving (1) from some more ultimate premise. This, of course, corresponds
to what is often regarded as a central tenet of Darwinism, namely the insis-
tence on the randomness (more properly, the nondirectedness) of variation.
Darwin himself was not quite so pure. The Origin allows for effects of "use
and disuse." Typically, this is supposed to work in tandem with natural se-
lection: organisms that do not use certain structures are more likely to gen-
erate offspring that lack those structures (tuco-tucos, burrowing rodents, are,
it is claimed, frequently blind; see Origin 137), and, since the deficient or-
ganisms do not squander resources on developing a useless characteristic,
they are at an advantage in the struggle for existence.40'41
Commitment to a stronger conception of Darwinian history makes it pos-
sible to answer questions, such as those which involve "perfections of struc-
ture," which lie beyond the scope of minimal Darwinian histories. This
commitment may be undertaken more or less pluralistically. That is, one may
allow as equally appropriate a number of different patterns for deriving
changes in trait frequencies, or one may insist that a particular style of ex-
planation should predominate. So, for example, Darwin's suggestion that
natural selection is the major agent of evolutionary change can be interpreted
as a commitment to preferring to understand the distribution of characteristics
in a group of organisms by invoking SIMPLE INDIVIDUAL SELECTION. Or, more
moderately, one can allow that the presence of many traits is to be explained
by employing CONSTRAINED INDIVIDUAL SELECTION Or one may decrease the
scope of selectionist explanations, by suggesting that the prevalence of some
40. Here Darwin's thinking seems to owe a debt to Geoffroy, who had insisted on
development as a process in which a limited stock of resources is expended on generating
adult morphology. In the first edition of the Origin, Darwin is quite cautious about other
systematic explanations of the presence of variations. While he is inclined to allow that
alterations in the environment are likely to increase variability (see, for example, Origin
134) he does not give much weight to the idea that the environment causes preferred types
of variation (see Origin 131-134 for a restrained discussion).
41. Darwin's claims for the dominance of natural selection as an agent of evolutionary
change are tempered by his frequent references to theunknownlaws of correlation and
balance. It is not always required to explain the presence of a trait by identifying the
(hypothetical) selective advantage that it conferred as it spread through ancestral popu-
lations. For the characteristic may be prevalent in a group because it is correlated with
another property whose presence is independently explainedperhaps by invoking natural
selection. Thus another Darwinian patternCONSTRAINED INDIDIVUAL SELECTIONeffec-
tively applies Darwin's selectionist patterns to suites of traits, allowing that such suites may
contain characteristics that are individually opposed by selection.
30 The Advancement of Science
(even many) traits is to be accounted for in some other wayperhaps in terms
of environmental effects, or use and disuse, or chance factors.
The Origin not only allows for the use of more or less ambitious notions
of Darwinian history, but also covers a range of positions on the priority of
selectionist explanations. This conclusion underscores a claim made by Gould
and Lewontin, who note that "the master's voice" is often more tolerant of
alternatives than is usually thought.42
So far I have indicated a number of different proposals that might be
reconstructed from the Origin. These differ first in whether they attempt to
explain changes in property frequencies along a lineage, and second in the
forms of explanation that they admit or to which they give emphasis.43 Minimal
Darwinism triumphed relatively swiftly in the 1860s. The next sections analyze
the nature of that triumph and the reasons for it.
5. Darwinian Practices
After the success of the Origin there are many changes in the ways in which
biology is done. Most obviously, there are modifications of the sets of state-
ments to which those who are counted as experts on the nature of living things
would give their assent. There are also changes in biological language: the
concept of species and the distinction between homologies and analogies are
altered by Darwin's work. As I have been emphasizing in the last two sections,
the questions that are viewed as important for biology are changed. New
problems come to occupy biological attention, old issues disappear. Further-
more, the changes in questions are accompanied by changes in the sets of
schemata that specify the ideal forms of answers.
However, these aspects, the alterations of belief and language that are
42. (Gould and Lewontin 1979). While I think that Gould and Lewontin are correct
in their claim that Darwin is less selectionist than he is often interpreted as being, Mayr is
right in his (1983) to point out that many of the alternative agents of evolutionary change
discussed by Darwin are now widely discredited. What does remainand this is sufficient
for Gould and Lewontin to make an important caseis the variety of factors that Darwin
would have lumped under "correlation and balance."
It is common for commentators to remark, parenthetically, that Darwin's commitment
to pluralism increases in successive editions of the Origin. I owe to Malcolm Kottler the
point that this quick assessment is misleading. Line-by-line comparison of the editions
shows Darwin inserting disjunctions of possible causes of evolutionary changes where he
had previously appealed to natural selection alone, but there is no addition of new mech-
anisms in later editions.
43. This does not exhaust the variety of versions of Darwinism. Nothing I have said
recognizes Darwin's commitment to evolutionary gradualism, nor have I allowed for a
possible Darwinian flirtation with selection of groups rather than individuals. Both these
further variants can be accommodated within the framework I have proposed, the one by
subjecting all Darwin's patterns to a gradualist constraint, the other by allowing for ana-
logues of his selectionist patterns that take groups and group properties as targets of
selection.
Darwin's Achievement 31
relatively familiar and the modification of problems and of explanatory pat-
terns (which I have seen as central to the Origin), do not exhaust the novelties
of Darwin's work. The Origin advances new ideas in methodology, ideas
about what kinds of characteristics are important in successful theories. It
suggests that certain types of observations and experiments are relevant to
biology, and it allows epistemic authority to people who have not hitherto
been counted as part of the community of scientists.
It will be useful to have a term to cover the disparate elements that were
altered by the Origin. Let us say that the practice of an individual scientist is
a multidimensional entity whose components are the language used in the
scientist's research, the statements about nature that the scientist accepts, the
questions that are counted as important, the schemata that are accepted for
answering questions (together with assessments of how they are applied to
"paradigm" examples and of how frequently they are likely to be exemplified
in coping with unsolved problems), the methodological views that are specific
to the research of the field of science, the canons of good observation and
experiment, and the standards for assessing the reliability of others (together
with particular views about the authority and reliability of contemporary co-
workers in the field). Before Darwin, those who dedicated themselves to the
study of living things differed widely in their practices. Not only were the
practices of individual researchers different from one another, but there were
groups of biologists whose practices had little in common. Champions of
teleology differed in numerous respects from the Naturphilosophen and from
compromisers like Owen. Even after the impact of the Origin, individual
biologists still differed in their practices. Yet, although the practices are di-
verse, almost all of them are Darwinian.
The ability to appreciate both the diversity of individual practices and the
unity of ideas that emerges at the level of the community will be crucial to
the argument of this book. Thus, besides the collection of individual practices
espoused by the individual scientists of the second half of the nineteenth
century, each affected in different ways by reading the Origin, there is a
consensus practice, something that represents the common elements of the
individual practices and that becomes part of the cross-generational system
of transmission of scientific ideas. In the latter nineteenth century, aspiring
naturalists acquire, as part of their training, a practice that involves a Dar-
winian core. To understand the change wrought by the Origin we need to
recognize both the weaker commitments to Darwinian themes that become
integrated into consensus practice and the more ambitious versions of Dar-
winism that are taken up by individuals.
Darwin modified the language of biology in two important ways. First, a
conceptualization of species that had been commonplace among his prede-
cessors was discarded: the language of late nineteenth century biology no
longer allows one to pin down the reference of species by declaring that "a
species comprehends all the individuals which descend from each other, or
from a common parentage, and those which resemble them as much as they
32 The Advancement of Science
do each other."44 But if he abandoned old ways of saying what species are,
Darwin and his followers largely agreed with the divisions of organisms into
species that had been favored by his predecessors. Thus, although there was
a shift in the way in which the referent of the term species was fixed (or, more
colloquially, how one explains what species are), the term retained its old
referent (the set of species taxa after Darwin is the same).45 Similarly, there
is continuity of reference of the term 'homology' (and its correlative, 'anal-
ogy'): Darwin calls homologies just those things that his predecessors have
picked out as homologies. Here, however, his writings implicitly suggest a
new descriptive account, one that would fix the referent of homology by
declaring that a homology is a characteristic found in two descendant species
as a consequence of their descent from a common ancestor.46
Darwin did effect large changes in the set of statements belonging to
consensus practice. His work introduced a variety of new claims about the
histories of particular organisms. The Origin is short on controversial new
generalizations, but it is a hodgepodge of specific original theses about bar-
nacles, pigeons, South American mammals, social insects, arctic flowers,
Scotch fir, and so forth.47
As I emphasized in Sections 3 and 4, Darwin's primary achievement lies
in introducing schemata for answering certain families of biological questions
and identifying the questions that biologists should address. The mass of
details is a cornucopia of illustrations. Consensus practice after Darwin ab-
sorbs the most striking examples from the Origin: every serious naturalist
now has to know about the South American mammals and the distribution
of arctic flora. The main alteration in the accepted statements of consensus
practice thus consists in the incorporation of those descriptions that best
illustrate the Darwinian schemata: these become the training paradigms for
subsequent generations. Of course, as further instances of Darwin's patterns
are generated and accepted, new examples may come to fulfil this pedagogical
function: melanism in moths comes to be the example of natural selection.
44. (Cuvier 1813 120, as quoted in Beatty 1985 267). The account I offer here is a
simplified version of Beatty's informative analysis.
45. The account of conceptual shifts suggested here will be developed in far more detail
in chapter 4. Beatty (1985) provides evidence for the continuity of reference of 'species.'
46. As Mayr notes (1982 465), this analysis is implicit in Darwin and allows a descriptive
way of fixing the distinction that Owen had drawn. The changes just mentioned, which are
the principal conceptual shifts attendant on the debate inspired by the Origin, are epis-
temologically unproblematic. They do not threaten to break down the communication
between Darwinians and their opponents, or even to create a situation in which "com-
munication . . . is inevitably partial" (Kuhn 1962/70 149). Partly because they were talking
about the same groups as species and the same traits as homologous, partly because the
loci of disagreement were so readily identifiable, Darwin and his critics had not the slightest
difficulty in communicating their ideas to one another.
47. Owen complained (Hull 1973b 170) that Darwin had introduced few "new facts."
In one sense, this is quite correct, for Darwin often takes the phenomena laboriously
assembled by others and makes historical claims about the organisms involved. After the
acceptance of the idea that all organisms are related by descent with modification, these
historical claims become new phenomena for subsequent biological research.
Darwin's Achievement 33
Meanwhile, individual scientists, concerned with specific biological questions
or particular groups, have a rich body of more esoteric lore, itself formulated
in Darwin's historical terms.
The introduction of the new schemata sets new questions for biology, in
that, after Darwin, naturalists are given the tasks of (i) finding instantiations
of the Darwinian schemata (that is, developing Darwinian explanations of
particular biological phenomena), (ii) finding ways of testing the hypotheses
that are put forward in instantiating Darwinian schemata, (iii) developing
theoretical accounts of the processes that are presupposed in Darwinian his-
tories (specifically such processes as hereditary transmission, and the origi-
nation and maintenance of variation). These tasks arise in different ways.
The first, (i), is simply the result of Darwin's claiming of the questions about
distribution, adaptation, and relationships, as legitimate and central questions
for biology, to be tackled by giving explanations that exemplify his patterns.
In attempting to instantiate the Darwinian schemata, biologists are compelled
to advance hypotheses about the historical development of life, and it is
incumbent on them to specify ways of testing these hypotheses (and hence
to undertake (ii)), if they are to avoid the charge that evolutionary biology
is simply an exercise in fantasizing.48 Finally, (iii) stems from recognition of
the fact that Darwin's theory is not only open-ended in provoking many
specific inquiries into the properties, relationships, and distributions of par-
ticular organisms, but also in raising very general questions about the historical
processes through which organisms have become modified.49
Although they are less central to the changes that Darwin recommends,
the Origin also offers new ideas about the kinds of information that are
relevant to biology, the ways in which that information can be acquired, and
those who can be trusted to supply such information. Darwin suggests ex-
perimental projects that will be important in formulating Darwinian histories.
Biogeographical accounts, for example, depend on assumptions about the
dispersal powers of organisms, and these can be tested in various ways. So
we find Darwin reporting his findings about the ability of various kinds of
seeds to survive immersion in salt water (Origin 358-361) and describing an
experiment on the survival of "just hatched molluscs" on a duck's foot sus-
pended in air (Origin 385). Furthermore, tests of the variation that is latent
in natural populations and of the powers of selection can draw on the craft
48. This line of criticism has surrounded Darwin's program since the beginning. It was
articulated by several of Darwin's most astute early criticsparticularly by Jules Pictet,
William Hopkins, and Fleeming Jenkin (see Hull 1973b). More or less sweeping versions
have been offered by contemporary critics: see (Ho and Saunders 1984) for a general
indictment of neo-Darwinism along these lines, and (Gould and Lewontin 1979) for an
attack on adaptationism. I have tried to disentangle what is correct about the criticisms
from various types of overstatements in several places (Kitcher 1985a chapters 2 and 7,
1985b section VII, 1988),
49. Darwin was very clear about the open-ended character of his theory, and about its
potential to inspire new kinds of scientific investigation. Not only is the last chapter of the
Origin self-consciously prophetic, but Darwin's letters also indicate his hopes for the future
development of biology (see, for example, F. Darwin 1888 II 128).
34 The Advancement of Science
knowledge of plant and animal breeders. In using the reports of pigeon fan-
ciers and beekeepers, Darwin was enlarging the circle of those considered
reliable informants on scientific matters. Aware of the problem of establishing
expertise, Darwin engaged in elaborate procedures to "calibrate" his potential
sources.50
Finally, Darwin's writings make a methodological claim that was not
widely appreciated in his own day and that continues to be questioned by his
critics today. As we shall see in the next section, the ideas of the Origin were
defended by stressing their ability to unify the phenomena. Time and again,
Darwin's detractors insisted that this defense was inadequate, that the kind
of argument it offered was at odds with the canons of good science. In reply,
Darwin insisted that many well-respected theorieshe liked to cite Maxwell's
electromagnetic theory, and it is clear that Lyell's geological theory also
furnished another, perhaps more immediate, examplepostulated entities
and processes that were not amenable to human observation (because they
were too small, were too remote, or took too long), and that these entities
and processes could, nonetheless, be known to exist because of the traces
they leave in the observable world of the present.
Darwin thus changed the consensus practice of biology along a number
of different dimensions. The central modification concerns the problems of
biology and the ideal forms of answers to them. Attendant on this are alter-
ations of the set of accepted statements and of the language of biology, as
well as changes in views of observational and experimental evidence and the
appropriate sources of that evidence. Finally, implicit in the Origin, and
explicit in some of Darwin's responses to reviews, is a methodological thesis
about the adequacy of certain kinds of scientific theories. To explore Darwin's
commitments here is to trespass on the topic of his argument for his proposed
modifications of biology, and it is to this topic that we now turn.
6. Darwin's Argument
Darwin claimed that the Origin is "one long argument." We have explored
the changes in practice that that argument was intended to support. Was
Darwin's reasoning up to the job?
Up to what job? There is no single powerful individual whose attitudes
are automatically those of the nineteenth century biological community, no
51. The reconstruction I give here is a philosophical elaboration of a scheme for in-
terpreting the reasoning of the Origin presented by Huxley (1896 72) and articulated in an
illuminating review article by Jonathan Hodge (1977).
36 The Advancement of Science
Lyell to keep his mind open until reading the "latter chapters which are the
most important of all on the favourable side" (F. Darwin 1888 II 166-167).
His approach is to marshall an impressive array of puzzling cases of geo-
graphical distribution, affinity of organisms, adaptation, and so forth, aiming
to convince his readers that there are numerous questions to which answers
fitting his schemata would bring welcome relief. Consider, for example, Dar-
win's partial agenda for biogeography. After describing the "American type
of structure" found in the birds and rodents of South America, Darwin sug-
gests that biologists ought to ask what has produced this common structure
(or, in the terms of Section 2, why neighboring centers of creation continually
bring forth embodiments of the same archetype). The similarities are too
numerous just to be dismissed as beyond the province of scientific explanation.
We see in these facts some deep organic bond, prevailing throughout space
and time, over the same areas of land and water, and independent of their
physical conditions. The naturalist must feel very little curiosity, who is not
led to inquire what this bond is.
This bond, on my theory, is simply inheritance, that cause which alone,
as far as we know, produces organisms quite like, or, as we see in the case
of varieties, nearly like each other. The dissimilarity of the inhabitants of
different regions may be attributed to modification through natural selection,
and in a quite subordinate degree to the direct influence of physical condi-
tions. (Origin 349-350)
This passage, brilliantly crafted to disarm objections in ways that I shall discuss
later, expresses a clear message. There are many details about particular
organisms that cry out for explanation, and if biologists commit themselves
to instantiating Darwin's schemata, then they will account for these otherwise
inexplicable details.52
However, consideration of cases raises skeptical doubts to the effect that
there are limits to the applicability of Darwin's schemata, so that his explan-
atory proposals are, at best, incomplete, or worse, incorrect. To construct a
Darwinian history will typically involve scientists in advancing hypotheses
about the existence of certain ancestral forms with particular properties. In
many cases, the fossil record will contain no remnants of such organisms.
How is this embarrassing lack of evidence to be understood?53 Moreover,
there are some properties of organisms which, in their final form, obviously
assist their bearers, but which would appear to be at best useless if they were
present in an incomplete state. Furthermore, Darwin's branching conception
of the history of life presupposes the possibility that the descendants of a
common ancestor may divide into two (or many more) groups whose members
are not interfertile. In all these cases, well-known findings in biology and
52. For other important passages that deliver the same message, see Origin 318-319,
339-341, 394, 440-444, 452-453, 471-480.
53. Various forms of the poverty of the fossil record are posed forcefully at Origin 280-
281, 287-288, 292, 301-303. One of Darwin's critics, Thomas Vernon Wollaston, saw the
state of the fossil record as being "the gravest of all objections" to Darwin's theory, but
he noted Darwin's frankness in admitting the facts (Hull 1973b/136).
Darwin's Achievement 37
geology seem to show that there are important and pervasive features of life
that both raise the kinds of questions he intends biology to address and fall
outside the range of his schemata.
Aware of the threat of limitation of his enterprise, Darwin devotes the
third part of his argument to an attempt to show that the difficulties are only
apparent. He offers an account of fossilization that is designed to explain why
many of the personae in the history of life leave no traces in the record. He
tries to show how complex structuressuch as eyesmay emerge under
selection, either because their incipient forms possess rudimentary advantages
(perhaps, though not necessarily, the same kinds of advantages as those
enjoyed by the final versions) or because they are correlated with advanta-
geous traits. He gives a careful survey of the phenomena surrounding sterility
and sketches a selectionist account of the emergence of a sterility barrier. By
arguing that apparently large problems dissolve under close scrutiny, Darwin
is able to defend the broad claim that the entire families of questions that he
proposes to make central to biology can be answered in the ways he suggests.
Let us now suppose that Darwin's contemporary naturalists appreciated
this chain of reasoning, and that they amended their practices in the following
way. First, they accepted Darwin's delineation of the important problems of
biology, accepting that questions of biogeography and comparative mor-
phology should be addressed by instantiating the patterns of Section 4.54 In
consequence of their initial commitments to Darwinian explanations in bio-
geography and comparative morphology, all of them abandoned their pre-
vious general belief in the fixity of species and all admitted that natural
selection is a possible agent of evolutionary change. Again, because of their
initial commitments, all espoused Darwin's reconceptualizations of species
and of homologies. Finally, all accepted the importance of the experimental
and observational projects that the Origin recommends, all recognized the
importance of the information on natural variation that breeders can supply,
and all hailed as a major new task for theoretical biology the project of
obtaining deeper insights into the phenomena of variation and heredity. This
modification of their individual practices is the acceptance of what I have
called minimal Darwinism.
Did the Origin provide these naturalists with good reasons for accepting
minimal Darwinism? Yes. The attitude concerning questions and schemata
for addressing them that I have supposed to be common to all naturalists,
together with the conception that natural selection is a possible agent of
evolutionary change, is well supported by the argument of the Origin. More-
over, once that attitude is in present, the changes elaborated in the latter part
of the last paragraph follow automatically: the conception of species must be
revised, experiments to determine dispersal powers become significant, the-
54. I shall assume that some, but by no means all, of them also favored the idea of
using SIMPLE INDIVIDUAL SELECTION (and other selectionist patterns) to address questions
about adaptation (and the modification of characteristics generally). Those who did so went
beyond minimal Darwinism to a more thorough involvement with Darwin's ideas.
38 The Advancement of Science
oretical questions about heredity and variation become major issues, and so
forth. My claim, then, is that naturalists living in Britain in 1859, starting
from any of the biological practices then popular, ought to have modified
their practice to accept minimal Darwinism, in light of the reasoning that I
have reconstructed from the Origin.
The best way to defend this claim is to see why a few people did not make
the shift to minimal Darwinism, and to show how Darwin and his supporters
had good responses to their objections. How did the critics attack the "long
argument"? What replies did Darwin, Huxley, Hooker, Gray, and others
make?
One important protest was that Darwin had no evidence of large-scale
modifications by natural or artificial selection, so that there was no good
reason to think that species could evolve in the way that he suggested. Typical
were the comments of the entomologist Thomas Vernon Wollaston:
there is no reason why varieties, strictly so called, . . . and also geographical
"sub-species," may not be brought about, even as a general rule, by this
process of "natural selection": but this, unfortunately, expresses the limits
between which we can imagine the law to operate, and which any evidence,
fairly deduced from facts, would seem to justify it: it is Mr. Darwin's fault
that he presses his theory too far.55
Because Darwin could only suggest the possibility of unlimited modifi-
cation, he was roundly chided by his critics for deserting the true path of
science. Drawing an invidious contrast, the applied mathematician William
Hopkins lauded the accomplishments of the physicists:
They are not content to say that it may be so, and thus to build up theories
based on bare possibilities. They prove, on the contrary, by modes of in-
vestigation that cannot be wrong, that phenomena exactly such as are ob-
served would necessarily, not by some vague possibility, result from the causes
hypothetically assigned, thus demonstrating those causes to be the true
causes. (Hull 1973b 239)
In a letter to Asa Gray, Darwin explained how Hopkins had failed to
appreciate the force of his argument:
I believe that Hopkins is so much opposed because his course of study has
never led him to reflect much on such subjects as geographical distribution,
homologies, &c., so that he does not feel it a relief to have some kind of
explanation. (F. Darwin 1888 II 237)
Although Darwin took some trouble in the Origin (and in his later Vari-
ation of Plants and Animals under Domestication) to show that artificial se-
lection is capable of producing quite dramatic modifications of organisms, his
principal response to the charge that variation is only limited is to shift the
55. (Hull 1973b 131). For similar remarks by Wollaston, Francois Jules Pictet, Samuel
Haughton, William Hopkins, and Fleeming Jenkin, see (Hull 1973b 135, 145, 224, 253,
304ff.).
Darwin's Achievement 39
burden of evidence. The analogy with artificial selection is not intended to
demonstratenor does it need to demonstratethat variation is unlimited.
Unless some reason can be given for supposing that there are limits to vari-
ation, then the explanatory power of the hypotheses that attribute descent
with modification justifies us in accepting them, even though modifications
as extensive as those which have been hypothesized have not been directly
observed. Only someone insensitive to the ability of the novel theory to
provide an account of phenomena that had hitherto seemed inexplicable (even
arbitrary)a nonbiologist like Hopkins, for examplewill fail to realize that
there is evidence for supposing that selection has quite extensive powers whose
action cannot be directly demonstrated. The opening chapters of the Origin
thus clear some space within which Darwin can defend his schemata for
tackling biological questions by appealing to their power to unify the
phenomena.56
Very well. But isn't there still some justice in the charge that "Mr. Darwin
presses his theory too far"? The charge is presented forcefully in Richard
Owen's version:
We have searched in vain, from Demaillet to Darwin, for the evidence or
the proof, that it is only necessary for one individual to vary, be it ever so
little, in order to [be led to] the conclusion that the variability is progressive
and unlimited, so as, in the course of generations, to change the species, the
genus, the order, or the class. We have no objection to this result of "natural
selection" in the abstract; but we desire to have reason for our faith. What
we do object to is, that science should be compromised through the as-
sumption of its true character by mere hypotheses, the logical consequences
of which are of such deep importance. (Hull 1973b 201)
Both Owen and Wollaston view Darwin as accepting the proposition that
the variation along a lineage exceeds any given limit, provided that sufficiently
many generations are considered, on the basis of the observation that a small
amount of variation can be observed over a small number of generations.
They see this evidence as compatible with the proposition that variation only
occurs within limits (say those marked by species boundaries) or as compatible
with the proposition that Darwin accepts. Darwin, then, is jumping to con-
clusions prematurely, recommending that "mere hypotheses" assume the
"true character" of science.
Darwin has a three-point reply to this criticism, a reply that is compressed
in the discussion of the South American fauna that I have quoted (Origin
349-50; see p. 36). The first point is to emphasize the length of time available,
and to draw (explicitly or implicitly) an analogy with geology. Natural selec-
56. Darwin's main argument stresses the unifying power of his schemata, but he cannot
resist giving subsidiary arguments. So, for example, the early chapters of the Origin cam-
paign against the idea that there is a natural boundary around species. This subsidiary
argument becomes very important to certain versions of Darwinismfor example, those
which take a nominalistic approach to species or those that emphasize evolutionary grad-
ualism. However, it is incidental to the more cautious version of Darwinism that was widely
accepted by British naturalists, the version of Darwinism with which I am concerned here.
40 The Advancement of Science
tion can be observed to produce small effects, just as the forces invoked by
Lyell can be seen to make minor changes in the features of our planet, and
we should no more expect to observe selection transcending a species bound-
ary than we should anticipate being able to observe erosion leveling a moun-
tain range. The second point is to emphasize the explanatory power of the
theory and the perplexity that we ought to feel when confronted with the
phenomenanaturalists who do not wonder about the relationships among
the South American animals must "feel very little curiosity." The third point,
directed at the prior practices of his contemporary British (and, to a lesser
extent, American and European) naturalists, is that, on anyone's account of
the phenomena there has been some process by which new species have been
called into being. Owen and Wollaston prefer not to speculate on what this
process is. It is not for them to judge whether the members of the new species
are the descendants of the old or whether there has been an entirely new
creation. But, Darwin argues, these two accounts are not equally well sup-
ported when we discover a "bond" among the species living in the same
region at different epochs. We know nothing of the process of special creation
and have no reasons for believing that the products of special creations should
be bound by the organic similarities observed (for example) in the American
fauna. We do know one way in which resemblance among organisms is main-
tained"that cause which alone, as far as we know, produces organisms quite
like, or, as we see in the case of varieties, nearly like each other" (my italics)
to wit, through inheritance. Owen and Wollaston are thus cast in the role of
seeing an account of species formation in terms of an unknown process sub-
jected to arbitrary constraints as equally credible as the suggestion that a
known process has been at work and that its effects are magnified on a time
scale that greatly exceeds human observational powers.
But Darwin's argument can be attacked from another direction. Should
his account be praised for its explanatory promise, or does it actually deliver
explanations? Darwin was sometimes inclined to make the stronger claim:
Thus, on the theory of descent with modification, the main facts with respect
to the mutual affinities of the extinct forms of life to each other and to living
forms, seem to me explained in a satisfactory manner. And they are wholly
inexplicable on any other view. (Origin 333)
Some reviewers were unconvinced. Hopkins lectured:
A phenomenon is properly said to be explained, more or less perfectly, when
it can be proved to be the necessary consequent of preceding phenomena,
or more especially, when it can be clearly referred to some recognised cause;
and any theory which enables us to do this may be said in a precise and
logical sense, to explain the phenomenon in question. But Mr. Darwin's
theory can explain nothing in this sense, because it cannot possibly assign
any necessary relation between the phenomena and the causes to which it
refers them. (Hull 1973b 267)
Hopkins's remarks make it clear that he regards Darwin's "explanations"
as falling short in two main respects: the hypotheses about descent with
Darwin's Achievement 41
those who do not fully appreciate the reasons are irrational. Darwin, I con-
tend, saw the issues more clearly than any of his opponents. But it is surely
true that there were many converts to Darwinism whose cognitive achieve-
ments were inferior to those of intelligent critics, who would have been unable
to reply to important objections that they did not perceive. Hopkins, Wol-
laston, Haughton, Owen, and, especially, Fleeming Jenkin engaged in rea-
soning of considerable sophistication, even though they did not appreciate
the possibility of Darwinian answers to their challenges. Labeling them as
"irrational" is somewhat akin to calling the Empire State Building "short,"
simply because it is no longer the tallest structure in town. The debate about
the Origin brought into ever sharper focus the structure of its central argu-
ment: Darwin's critics (and his champions) played an important constructive
role in making it possible to appreciate the reasoning I have outlined here.
7. The Synthesis
After sixty years of acceptance of minimal Darwinism and serious debate
about the hegemony of natural selection, the modern synthesis of the 1930s
brought Darwin's ambitious proposals into the consensus practice of biolo-
gists. To understand the synthesis and how it extended the Darwinian ideas
I have analyzed, it is convenient to divide it into two stages. First comes the
mathematical work, the definitive reconciliation of Darwin and Mendel and
the elaboration of theoretical population genetics.57
Mathematical population genetics takes as its central question the general
problem, Given a population P with characteristics W and subject to an initial
genetic distribution D0, what will be the (expected) genetic distribution after
n generations? In formulating concrete instances of the problem, a number
of different kinds of features of the population must be specified: we must
state whether the population is (effectively) infinite, whether the organisms
reproduce sexually, whether the organisms mate at random. We must also
identify the "forces" acting on the populationmutation, migration, natural
selectionand offer a genetic description that lists the loci involved, the
number of alleles at each locus, the linkage relations (if any), any disruptions
of normal meiosis, and so forth. The very simplest case for a sexually repro-
ducing organism is to suppose that we have an infinite, panmictic (randomly
mating) population, with no mutation and no selection, in which we are tracing
the frequencies of genes and binary gene combinations at a single locus, a
locus at which there are two alternative alleles. In this case, it is easy to prove
(as Hardy and Weinberg independently did in 1908) that if the relative fre-
quencies of the alleles A and a are p and q (where p + q = 1), then whatever
the initial frequencies of the genotypes (here the binary gene combinations:
57. That work began well before 1930. Its roots are in the neglected writings of Yule
and in the independent discoveries of Hardy and Weinberg, and its first flowering is in a
pioneering paper by Fisher (1918).
Darwin's Achievement 43
AA, Aa, ad) the population will move in one generation to an equilibrium
in which the distribution of genotypes is p2 AA, 2pq Aa, q2 aa. Population
genetics tackles the general problem by employing more intricate (often far
more intricate) versions of the same type of combinatorial reasoning that is
used in this simple case.
Selection is represented by assigning to genetic combinations coefficients
that summarize their probabilities of making contributions to the next gen-
eration. We can fix ideas by thinking only in terms of survival. Selection may
act upon a population because organisms bearing some genetic combinations
are more likely to survive to maturity than others are. Assessing the trans-
mission of alleles into the next generation, we shall need to take the survival
probabilities into account. So, for example, if we are investigating an infinite
panmictic population with two alleles at a locus, our original distribution of
genotypes may be
58. Here, and in what follows, square brackets enclose alternative completions of a
sentence.
59. If the population starts from a Hardy-Weinberg equilibrium, then the F's can be
computed from the p's. However, the general treatment that descends from Fisher, Hal-
dane, and Wright does not assume that this is necessarily so.
60. It is not required that the w's be constants. Indeed the values of the fitnesses might
depend on the frequencies either of alleles or of allelic combinations within P. Thus the
schema allows for frequency-dependent selection.
61. Here one could also insert a clause allowing for the possibility of meiotic drive:
(f) If an organism has the genetic combination aijaik at the ith locus, then the probability
that the chromosome containing the allele aij will be transmitted to a gamete is gijk {spec-
ification for all pairwise combinations at all loci}.
It would be anachronistic to include this as part of the schema devised by Fisher,
Haldane, and Wright, but it should be incorporated in the present version of the pattern
for solving trajectory problems in population genetics. For discussion of meiotic drive, see
(Crow 1979).
Darwin's Achievement 45
62. In the case of random mating the g's vary directly as the products of the relative
frequencies of the allelic combinations.
63. Instead of treating Fisher, Haldane, and Wright as introducing a general schema
into population genetics (or, more exactly, founding the subject by proposing a schema)
one could view them as specifying a general model type. The subsequent work of mathe-
matical population geneticists would then be regarded as the specification of some of the
features that are left open in the general model type, so as to yield particular population
genetics models. This approach to the subject would result in a picture akin to that elab-
orated by Richard Lewontin in chapter 1 of his (1974) and articulated by Elisabeth Lloyd
(1984, 1988). For certain purposes, this suggestion would be equivalent to the approach I
pursue here. However, I am interested not only in the structure of the models but also in
the relation between the work of Fisher, Haldane, and Wright and the families of problems
addressed within evolutionary biology, and, for this endeavor, the focus on explanation
and strategies for responding to explanation-seeking questions seems more straightforward.
I would not be perturbed if Lloyd and others who favor a semantic conception of theories
find it possible to translate what I say into their own preferred idiom.
46 The Advancement of Science
NEO-DARWINIAN SELECTION
Question: Why is the distribution of properties P1 . . . , Pk in relative
frequencies r1 , . . . , rk (Eri = 1) found in group G?
Answer:
(1) Among the ancestors of G there is a group of contemporaneous
organisms, G0, such that (i) there was variation at n loci among members
of G0; (ii) at the ith locus there were mi alternative alleles present in the
G0-G sequence.
(2) In the environment common to all organisms in the G0-G sequence,
organisms with the allelic combination a11a11 . . . anlanl have probability
S1...1,jof having the trait Pj, . . . {specification of the gene-environment-
phenotype relations for all the allelic combinations}.64
(3) Analysis of the ecological conditions and the effects on their bearers
of P1, . . ., Pk.
Showing
(4) The fitness of the allelic combinationa11a11...anlanl is w1 . . . 1, the
64. Here I specifically allow that the forms of phenotypes may depend on aspects of
the environment that vary from organism to organism. In many instantiations of the se-
lectionist schema this possibility is ignored, and it is assumed that there is a uniform
association of underlying allelic combinations with phenotypic traits. My schema could be
complicated further by allowing for the possibility of changing environments. But, while I
believe that Fisher, Haldane, and Wright would all have recognized the possibility that
differences in microenvironment may make a difference to the phenotypic expression of
an underlying genotype, consideration of fluctuating common environments seems to lie
beyond the purview of their analyses.
Darwin's Achievement 47
65. (4) is derived from (2) and (3) by considering the probability that a bearer of an
allelic combination will manifest a phenotypic trait and using the ecological analysis to
compute the expected reproductive success of bearers of the trait.
66. Since there are mi alternative alleles at the ith locus in the sequence G 0 -G, this
requires that the number of alternative alleles at the ith locus in G0, mi*, be less than or
equal to mi. It is not required that mi* = mi, since we should allow that new alleles may
appear along the lineage.
48 The Advancement of Science
unknown) ecological analysis that would form part of an explanation of the actual distri-
bution of traits, and it seeks to uncover the strength of selection or to calibrate a proposed
means of measuring the intensity of selection.
69. This becomes especially obvious in retrospective analyses of the synthesissuch as
those offered in (Mayr and Provine 1980)where particular authors can readily be assigned
responsibility for particular fields. However, there is some overlap, and, in particular,
(Dobzhansky 1937) not only addresses the questions arising from genetics but also surveys
a number of other biological topics in less detail.
50 The Advancement of Science
became those of fathoming the genetic changes required for descendants from
a common ancestor to become reproductively isolated and of identifying which
exemplifications of NEO-DARWINIAN SELECTION would be most frequently em-
ployed in accounting for these changes. Mayr extended his analysis by pro-
posing that the attainment of reproductive isolation required the interruption
of gene flow (conceived of as a homogenizing agent), so that geographical
isolation is a necessary prior condition of reproductive isolation.
Perhaps the most difficult task was to integrate the new ideas about ge-
netics, populations, and the explanation of evolutionary change with the
phenomena of paleontology. Post-Darwinian studies of the fossil record had
frequently inspired doubts about the importance of natural selection.70 Simp-
son's accomplishment in his (1944) was to demonstrate that there was no
need to extend the resources of the explanatory schema articulated by Fisher,
Wright, Haldane, and Dobzhansky. NEO-DARWINIAN SELECTION would suffice
for the phenomena of macroevolution.
Part of what Simpson accomplished was a "consistency argument" (Gould
1980b 161). However, he also set a new agenda for paleontology by dem-
onstrating possibilities of measuring and comparing evolutionary rates and
using the results to arrive at conclusions about evolutionary modes. Paleon-
tology, in its new guise, could examine the effects of the microevolutionary
processes analyzed by Dobzhansky and his co-workers as they were repeated
over vast stretches of time.
70. (Bowler 1983) provides an extremely lucid and useful review of the paleontological
worries about natural selection, and the search for alternative mechanisms of evolutionary
change.
Darwin's Achievement 51
cesses. So, for example, the cases of industrial melanism in moths and of
cowbird-oropendula parasitism furnish the standard exemplars of the action
of natural selection in the wild. Contemporary biology thus enlarges the store
of claims about the particular features of particular types of organisms that
Darwin advanced in the Origin. We have greatly multiplied the phenomena
of biogeography, comparative anatomy, and adaptation. In addition, the ad-
vances of the synthesis have been extended in study after study of the genetic
variation in natural populations and of mutation, recombination, and selection
both in the laboratory and in the field. Moreover, as we shall see, the triumphs
of molecular biology have made it possible to add evolutionary phenomena
of a qualitatively different type: for example, we can discuss homologies not
only in terms of anatomical structures but by reference to protein composition
as well.
We cannot think of evolutionary biology, even at the level of phenomena,
as making progress through accumulation. Although a statement becomes
part of the practice of every member of a subcommunity of biologists, the
subcommunity deemed by the full community to be especially authoritative
with respect to statements of that type, it may nonetheless be abandoned at
a later stage. Consensus ideas about the dispersal of animals, including some
accounts of current distributions that had been universally accepted, had to
be revised in light of plate tectonics and its concomitant views of the former
relations of the continents. Similarly, new discoveries of fossil hominids have
forced paleontologists, primatologists, and physical anthropologists to scrap
previously well-entrenched conclusions about the divergence of the hominid
line from the great apes. Nonetheless, there is something apparently right
about the idea that we know far more about evolutionary relationships than
Darwin did. To capture that idea, I suggest we recognize that, while some
descriptions of phenomena are revised or discarded, many become stable
parts of consensus practice. The seeming growth in our understanding of
biogeography, paleontology, species diversity, and species relationships is, I
think, partly captured by the presence in later practices of an increasing
number of stable reports of phenomena, with the rate of increase greatly
outstripping the rate of revision.
73. I should note that there are levels of comparison intermediate between the gross
anatomical features and the fine structures of molecules. Advances in cytology have made
possible comparisons of chromosome number, and even, in some cases, of banding patterns.
The case of humans and chimpanzees provides a particularly fine example.
74. As Bowler notes (1983 141-146), the Darwinian penchant for fanciful tales about
benefits fueled the search for alternative agents of evolutionary change.
54 The Advancement of Science
75
"hardened." Part of this hardening became manifest in the attitude that it
is appropriate for a biologist, concerned to understand the prevalence of a
particular trait in a particular group, to start from the idea that that trait has
been separately fashioned by selection. The explanatory strategy of first re-
sort, then, is to instantiate NEO-DARWINIAN SELECTION in such a fashion that
only the selective value of the focal trait is taken into account and factors
besides selection are downplayed.
To a first approximation, the Darwinian community divides into adapta-
tionists and antiadaptationists. But this approximation is very crude indeed.
As with the debate about variation and its causes, the issue is ultimately about
frequencies. Everyone will concede that there are some instances in which
selection of traits cannot proceed independently: cases of pleiotropy, allom-
etry, and close linkage are well known. But, for some Darwinians, selection
is very powerful. We should expect, they claim, that modifier genes will evolve
to suppress unwanted side effects, that linkages can be broken, developmental
associations modified. In effect, we can start by taking the traits of an organism
to evolve separately unless or until we are forced to complicate our analysis.76
Others believe that developmental constraints are likely to be omnipresent,
and that we have little hope of arriving at adequate explanations of the
presence of traits unless we probe the mechanisms of ontogenesis. For my
purposes, the important point to note is that this debate, like that about
genetic variation, proceeds within the scope of commitment to those ideas I
have identified as central to Darwinism and neo-Darwinism.
75. The term is Gould's. See his (1983) for a perceptive analysis of the later writings
of the architects which shows how the subsequent editions of their major works give greater
emphasis to selection with concomitant downplaying of other ideas about the causes of
evolutionary change. The trend was at least partly motivated by the perception that selective
advantages could be found for what initially appeared to be nonfunctional traits. Particularly
pivotal was the study by Cain and Sheppard of band polymorphism on shells of the snail
Cepaea nemoralis, where it was shown how apparently trivial variations were correlated
with changes in the color of the substrate, and that coloration affected rates of predation.
See (Beatty 1986) for an account of how the selectionist explanation of these apparently
nonadaptive variations led to increasing confidence in the power of selection.
76. This attitude is most obvious in the writings of Richard Dawkins (1976, 1982, 1986).
Dawkins can legitimately claim, I think, to be elaborating with considerable clarity a view
of selection that stems from Fisher (1930). He also represents the thinking of a significant
number of evolutionary biologists whose strategies for dealing with particular examples
tacitly embody his ideas.
Darwin's Achievement 55
for the large-scale features of evolutionary history. Both he and Simpson were
initially drawn to the conception of population genetics articulated by Wright,
a conception in which population structure may play an important role and
in which drift in small populations is an important factor both in the process
of speciation and in the breakthrough into new adaptive zones. In the 1940s
and 1950s, Wright's ideas faded into the background. As the synthesis hard-
ened, paleontologists became more confident in the power of natural selec-
tion, they began to conceive macroevolutionary processes as extrapolations
of microevolutionary processes, and they offered exemplifications of NEO-
DARWINIAN SELECTION that were simply extensions of those offered in under-
standing changes in allele frequencies in contemporary populations.
This situation changed in the early 1970s with the publication of a seminal
paper (Eldredge and Gould 1972) which sparked the most publicized evo-
lutionary debate of recent times. According to Eldredge and Gould, there is
rapid evolutionary change (rapid relative to the geological time scale) while
speciation is occurring. Once a species has formed, however, its morphology
remains fairly constant. If Eldredge and Gould are correct, then the absence
of intermediates is to be expected (since the segments of the record in which
they would be found are small in comparison with the predominant periods
of stasis) and there is an interesting further question as to how rapid specia-
tion and prolonged stasis are to be understood in terms of NEO-DARWINIAN
SELECTION.
Although the theory of punctuated equilibria begins from a thesis about
the geometry of phylogenetic treestypical phylogenetic trees, for this dispute
is, like our previous controversies, concerned with the frequency of various
scenarios in the history of lifetwo larger points of controversy quickly
emerge. First, do the kinds of processes that occur in microevolutionary
change, processes that can be studied on living populations and that have
been studied ever since Dobzhansky's pioneering work, underlie the rapid
speciation events? Conceding that the changes that occur when evolutionary
stasis is punctuated are consistent with theoretical population genetics, some
defenders of punctuated equilibrium (notably Gould) believe that they are
importantly different from the familiar examples of microevolutionary change
disclosed in postsynthesis research. Gould's claim can be captured in our
terms by suggesting that instantiations of NEO-DARWINIAN SELECTION quite
different from those that figure in orthodox neo-Darwinian explanations will
be needed to understand the rapid speciation events (and, perhaps, even to
understand the periods of stasis in cases where there are environmental
changes).
A second major development of the theory of punctuated equilibrium is
the attempt to expand the class of neo-Darwinian explanations by invoking
higher-level selection. Specifically, Stanley, Gould, Eldredge, and Elizabeth
Vrba have all proposed that major features of the diversity of life should be
explained in terms of species-selection. If a species has a species-level property
(the tendency to have a particular population structure, for example) which
increases its propensity for leaving descendant species, and if this property
56 The Advancement of Science
is retained among the descendant species, then we can expect species with
the property to increase in frequency. We can regard this as a claim to the
effect that there is need for an expansion of neo-Darwinism by introducing
schemata analogous to Darwin's, that would be instantiated in accounting for
macroevolutionary phenomena.
1. For concreteness, one might suppose that the period of the cycle I envisage is a
few months. I believe that this idealization will suffice to capture the major features of
scientific change even in fast-moving disciplines.
2. It is useful to distinguish two different relations of authority and deference here.
First, there is the trust that is placed in the experts by other scientists and members of the
broader public, the trust exemplified in Dobzhansky's willingness to defer to Wright's
mathematics (Provine 1986 345-347) or the layperson's readiness to credit the views of
paleontologists about the character of Silurian life. Second, there are the varied propensities
58
The Microstructure of Scientific Change 59
The apprentices are trained by the veterans. I shall suppose that each of
them forms an initial individual practice that contains the consensus practice
as a part, but that each extends this in ways that depend partly on the particular
elaboration of consensus practice that is presented and partly on the cognitive
state that the apprentice brings to the training process. Because different
teachers elaborate consensus practice in different ways and because appren-
tices vary in their early intellectual ontogenies, there is variation in the in-
dividual practices with which they leave the training period and begin their
work. Moreover, apprentices join the scientific community with varying de-
grees of initial credibility, determined (in part) by the status of the veterans
from whom training is received.
During the work phase, individual practices (those of apprentices and
veterans alike) are modified through conversations with peers and through
encounters with nature. As information from others is accepted, modified,
extended, or rejected, so assignments of credibility change. Those whose
credibility declines sufficiently far are effectively excluded from further con-
versation, and they may drop out of the community altogether. Occasionally
outsiders, people who have not been trained by community veterans, enter
the community, earning credibility by advancing ideas that members of the
community endorse. At the end of the work phase, the modifications of
individual practices induce, according to rules that form part of the social
system of the community, a change in consensus practice. Those who endorse
this change may remain in the community to serve as veterans in the next
period (although some of them may retire). Those who do not are excluded
from the community and play no further role. The cycle thus begins anew
with a population of veterans and a new crop of apprentices.
The philosophical problems about scientific change can be regarded as
questions about iterations of this cycle. Descriptive problems are concerned
with clarifying the notions I have employed. Prescriptive problems focus on
the conditions under which iterations of the cycle yield a process that attains
or is likely to attainthe goals that we ascribe to the enterprise.
To provide a general descriptive account of scientific change we need a
clear view of the states and processes that occur as constituents of the cycle.
Cognitive States. Science is not done by logically omniscient lone knowers
but by biological systems with certain kinds of capacities and limitations. At
the most fine-grained level, scientific change involves modifications of the
cognitive states of limited biological systems. What are the characteristics of
these systems? What kinds of cognitive states can they be in? What are their
limitations? What types of transitions among their states are possible? What
types are debarred? What kinds of goals and interests do these systems have?
Practices. The cognitive histories of individuals are highly idiosyncratic.
of some scientists in one group to defer to others (either in the same group or in a different
specialty) with respect to issues that are not yet regarded as settled (not yet incorporated
into consensus practice). The sources and effects of this latter type of variation will be
studied in Chapter 8.
60 The Advancement of Science
Abstracting from some of the everyday flux, we can identify a cluster of
scientific commitments that remains relatively stable during a single period of
the cycle. These commitments, collectively the scientist's individual practice,
are manifested to others in her behavior: in her assent to certain statements,
her pursuit of certain questions, her use of certain instruments or techniques,
her production of texts with certain structures, her disposition to rely on some
informants and not on others. What are the components of an individual
practice? How should each component be understood?
Individual practices are still highly diverse. It is a fiction to suppose (con-
centrating on the simplest and most frequently acknowledged component of
practice) that there is some corpus of statements such that each individual
scientist has all members of the corpus and just those statements about the
relevant aspects of nature stored inside his head, ready for assent under
querying. Most scientists depend on tools and external aids (including other
scientists) and they have private stores of information (in lab notebooks and
on computers), their own very particular instruments and samples. But beyond
this individual variation, there is surely something that all members of the
scientific community, the expert group, share: consensus practice.
Consensus practice changes in response to modifications of individual
practices; individual practices alter as a result of changes in individuals' cog-
nitive states. What drives these latter changes? I shall start with an admittedly
rough distinction. Sometimes scientists modify their cognitive states as results
of asocial interactions, sometimes they change their minds through social
exchanges. The obvious exemplars for the former are the solitary experi-
mentalist at work with apparatus and samples and the lone field observer
attending to the organismsalthough I shall also take encounters with nature
to cover those occasions on which scientists reflect, constructing chains of
reasoning that modify their commitments. Paradigm cases of conversations
with peers are those episodes in which one scientist is told something by
another (and believes it) or when a change in commitment is caused by the
reading of a text. The point of the distinction is evidently to separate those
episodes that (very roughly) consist in finding things out for oneself from
those in which one relies on others, but two important caveats need to be
noted. First, there are many episodes in which cognitive states change in
response both to social exchanges and to asocial interactions: consider, for
example, occasions on which several scientists are using an intricate piece of
apparatus and relying on each other for confirmation of its proper functioning.
Perhaps such episodes can be decomposed into smaller incidents, each of
which can be happily classified in terms of my distinction. But, for my pur-
poses, there is no need to assume that this is so. I am primarily interested in
understanding the dynamics of cognitive change in response to social and
asocial sources. If exemplary and unproblematic instances can be handled,
the project of understanding how to compound the "forces" can be under-
taken later and applied case by case without any prior categorization of
complex episodes. Second, in differentiating those occasions on which sci-
entists "encounter nature," I have no intention of suggesting that the impact
The Microstructure of Scientific Change 61
of stimuli from the asocial world is independent of the subject's prior state,
including, very probably, her prior history of social interactions. The solitary
experimentalist may bring to the latest trial all kinds of commitments, many
of which were formed by relying on othersindeed, trust in the apparatus
may be founded in social interactions with people to whom the experimenter
defers. The point of selecting experiments as paradigm encounters with nature
is not that a stimulus from the asocial world is the unique determinant of
cognitive change but that it is one relevant causal factor.
Despite the important advances in our recent understanding of perception
and cognition, despite detailed studies in the history of science and thorough
investigations of some aspects of the sociology of science, we do not yet know
enough to give all the details about the facets of science to which I have
pointed. Nor do we yet know how much variation there is likely to be from
field to field and from time to time. Nevertheless, for some philosophical
questions, including those that concern me in this book, what we need are
not so much detailed answers as a sense of the range of possibilities. We have
to appreciate the options.
Our primary prescriptive tasks are to give an account of the goals of science
and to derive from it a theory of what constitutes progress in science, to
understand how individuals ought to behave and how their social relations
should be designed to facilitate attainment of the goals. Even without detailed
knowledge of all elements of the life cycle, we can address these issues,
provided only that we can replace Legend's austere sentential framework with
something which (for all its residual sketchiness) is closer tobut still an
idealization ofscientific practice.
2. Individual Cognition
Legend was often elaborated as though a simple view of mental activity was
presupposed. Think of the mind as a box whose contents are declarative
statements or propositions.3 At any given stage in the history of a field of
science, we can suppose that all (competent) workers in the field have exactly
the same propositions stored in their boxes. In response to new information
personal observation of some aspect of nature or reading of a text, such as
the Originthe contents of the boxes are modified and the housekeeping
apparatus, the embodiment of the scientists' logical and methodological pro-
clivities, goes to work to effect further reorganization, sometimes leading to
the acceptance of large theoretical claims. One task of philosophy of science,
traditionally conceived, is to understand the functioning of this housekeeping
machinery. This simple tale of mental life, which, I claim, is taken for granted
3. I shall use these terms interchangeably, without any great attention to philosophical
niceties. Issues about meaning, intention, truth bearers, and so forth, will only be relevant
to my concerns in this book when the languages of scientific practices are under discussion
(see Section 7 and Chapters 4 and 5). There my views will be made explicit.
62 The Advancement of Science
not only by the adherents of Legend but also by many of their philosophical
successors,4 is so far from any realistic psychological account of the thought
and behavior of scientistsas for example, in the responses to the Origin,
reviewed in the last chapterthat questions about the roles of reason, sub-
jectivity, and social forces cannot be satisfactorily posed, let alone tackled,
without replacing it.
Although minds are not simply receptacles for storing representations,
one facet of mental activity, prominent in the psychological lives of scientists,
is representation (and the processing of representations). Thus one question
that needs to be addressed concerns the representation of representations:
should we think in terms of propositions (statements), of images, or of some-
thing quite different? Another issue concerns the organization of represen-
tations: how are they stored, how activated? Lastly, we must decide what
other capacities, dispositions, faculties, to attribute to the scientific mind, and
how they interact with one another. By focusing on these problems, we obtain
a substitute for the simple propositions-in-a-box picture of mental life, some-
thing still highly idealized but (we may hope) sufficiently realistic to be useful
in understanding the microstructure of scientific change.
There is no current psychological consensus about the nature of repre-
sentations. Following one major approach in cognitive psychology and arti-
ficial intelligence, I shall assume that mental activity often involves the storage
and retrieval of propositions, and that we can think of such processes on the
5. Pioneering work on the role of imagery in cognitive tasks has been done by Roger
Shepard and his associates. See, in particular (Shepard and Chipman 1970, Shepard and
Metzler 1971). Stephen Kosslyn's (1981) is an important summary of the case for images,
and Ned Block's (1981) collects many of the important contributions to the debate. Chapter
12 of (Goldman 1986) provides a useful overview and defends what I regard as a sensible
ecumenical position.
6. I resist a much-debated current challenge to the propositional character of human
cognition (see, for example, Paul Churchland 1989). I do so because I do not yet see how
the envisaged alternativethinking of cognition in terms of "patterns of network activa-
tion"can do justice to the articulation of propositions and reasons that is so prominent
in the growth of scientific knowledge. The difficulties of Churchland's account are apparent
from his conflation of very different scientific notionsexplanations, theories, and percep-
tual classifications are all treated as if there were no differences among them.
64 The Advancement of Science
account for them, precisely because it treats the contents of the box as if they
were independent atoms of equal status.7
We can do some justice to the simple observations by drawing inspiration
from recent writings in artificial intelligence and cognitive psychology,8 and
by supposing that our minds are organized along the lines of the system
depicted in Figure 3.1. This system is enclosed within a boundary across which
information flows from the external environment and decisions are translated
into action. The central division within the system is between those items that
are activated, assumed to be transferred to working memory, and those that
are stored in an unactivated condition. The current state of activation, the
condition of working memory, may start a number of processes. Propositions
(or images) already present in working memory may be transferred to declar-
ative memory. Alternatively, the current state of working memory may initiate
a search process for some desired representation from declarative memory.
If the search is successful, the representation is transferred to working mem-
ory. In analogous fashion, currently pursued goals may be stored among the
long-term goals, or the current state of working memory may promote a search
for stored goals, which, if successful, issues in the retrieval of goals to working
memory.
Figure 3.1 shows a third storage system. Procedural memory includes the
skills, know-how, and inferential propensities that are stored in the system.
Many items in this subsystem are relevant in understanding scientific behavior.
Consider for example the performances of trained experimentalists or ob-
servers. Think of the abilities to manipulate pieces of apparatus, to make the
instruments work. There are also the inferential propensities of subjects.
Under appropriate cuing a subject may form a representation of a particular
problemwriting down the equation that governs a particular type of physical
system or conceiving of an evolutionary situation by invoking a particular
ecological schemaand then proceed to draw various conclusions. I shall
understand these kinds of performances to be grounded in the person's in-
ferential propensities.
This brief account of the subsystems and their contents does not specify
the manner in which the representations or propensities are stored. With
respect to the contents of working memory I shall suppose that representations
may be endorsed or merely entertained. So, to continue an obvious (but
fanciful) analogy, we might think of propositions as being entered in working
memory either in ink (endorsement) or in pencil (entertainment). Operations
on the contents of working memory can alter the mode of storage, promoting
the entertained or relegating the formerly endorsed. For the contents of the
long-term subsystems, the representations in declarative memory, the long-
term goals, the stored propensities, I shall suppose that there are important
differences in strength or valence. Valence varies directly with the probability
of activation: items with high valence are relatively accessible, those with low
valence inaccessible.
How seriously should we take this cognitive geography? It would be folly
to claim that it is corrector even approximately correctfor there is too
little evidence and too many serious unanswered questions. But I do take it
to have heuristic value, enabling us to pose clearly questions about the growth
of scientific knowledge and to recognize distinctions and problems that have
often been overlooked.9
3. Cognitive Constraints
Any conclusions about how cognitive systems ought to behave must be based
upon an understanding of their limitations. The most obvious limitations on
systems of the type I have described are those that result from considerations
of size. Working memory is apparently subject to fairly severe constraints
(Miller 1956), and although declarative memory, procedural memory, and
the set of long-term goals may be very large, there is evidence to suggest that
9. As I maintained in footnote 4, and as I elaborate in (Kitcher 1992), every episte-
mology needs a psychology. The advantage of making one's psychological picture relatively
explicit is that epistemological claims and concerns that are linked to some discredited
element of the picture can readily be abandoned. And, of course, there is always the
possibility of providing amusement for those who think that the picture is downright silly.
Perhaps amusement will issue in attempts to achieve something better.
66 The Advancement of Science
size limitations may affect how much of these subsystems can readily be
searched (see Anderson 1983, Goldman 1986 chapter 10, Loftus 1979, Hyde
and Jenkins 1973). Scientists facing complex decision problemspondering,
for example, whether the evidence is sufficiently strong to accept the main
claims of punctuated equilibriummust therefore find ways of organizing
their consideration of bodies of evidence that are too large to be surveyed at
once.
Size limitations can be relieved through the use of external aids, notebooks
and computer memories for example, but the use of these aids only makes
public the need to store information so that the scientist can gain access to
it. It is, I think, quite likely that many of the sources of a scientist's belief
are subsequently lost, in the sense that there is no possibility at a later time
for the scientist to retrace his original route to the belief from the traces that
endure in declarative memory, on paper, or inside his computer. If this is
correct, then the philosophical ideal of the permanent possibility of rechecking
may be unrealizable for an individual.10
More subtle constraints may result from the dynamics of the transactions
among subsystems. The mere fact that an inferential propensity may be pres-
ent in procedural memory does not entail that it will be activated under the
appropriate circumstances (Nisbett and Ross 1980, Kahneman, Slovic, and
Tversky 1982). Similarly, search operations through declarative memory may
be unsuccessful because the state of working memory cues the search in the
wrong way. The general problem here is that certain states of the system may
be inaccessible from certain other states. I shall develop this point by con-
sidering how it might bear on some issues in the philosophy of science.
The general problem underlies one version of the notorious issue of the
theory-ladenness of perception. We can formulate that issue as follows: Ex-
ternal objects participate in causal processes which impinge on the cognitive
subject: the subjectively proximal portion of such a process is the stimulus,
and I shall suppose that stimuli can be individuated in some physicalistically
unproblematic fashion. As a result of the presence of the stimulus and the
prior state of the cognitive system, conceived as the totality of contents of
the storage subsystems, the transactional processes among subsystems, and
the state of activation (i.e., the condition of working memory), the system
goes into a new state. If the new state of the system involves the insertion of
a new believed proposition into working memory (possibly, but by no means
necessarily, a proposition that was previously entertained in working mem-
ory) , then I shall call that proposition a perceptually induced belief (or a PI
belief).
The thesis that perception provides a theory-independent basis, intersub-
jectively available, for the checking and correction of theoretical claims can
10. Points akin to this have been made by Alvin Goldman (1986) and Gilbert Harman
(1987), both of whom question the traditional supposition that it is epistemically required
that people be able to trace the justifications for their beliefs by pointing out the difficulty
or impossibility of doing so.
The Microstructure of Scientific Change 67
easily be articulated. Suppose that the mechanisms leading from the boundary
of the system to working memory are functioning properly.11 Then, whatever
the cognitive state of the system, the presentation of the same stimulus will
lead to the incorporation of the same PI beliefs. Moreover, provided that the
system is in a properly functioning state (intuitively, that the subject is at-
tentive, not hallucinating, and so forth) and that the external conditions are
normal, then the PI beliefs will be true.
Numerous writers over the past two decades (Sellars 1956, Hanson 1958,
Kuhn 1962/70, Feyerabend 1965, Hesse 1970, Churchland 1979, 1989) have
denied this. The trained paleontologist, inspecting the strata exposed in a
railway or road cutting, sees the advance and retreat of Paleozoic oceans; the
novice sees only a rock face, traversed by slanting lines. Or, to revert to one
of the most cited examples, Johann Kepler and Tycho Brahe, watching the
dawn together, see, respectively, the sun coming into view as the earth turns
and the sun beginning its daily journey around a static earth. The perceptual
stimuli activate, either directly or through working memory, propensities that
the paleontologist has acquired by dint of long training and that the novice
lacks. The common stimulus that impinges on Kepler and Tycho activates
different propensities in the two men because of the character of their back-
ground cognitive states, specifically because of their commitments to different
systems of astronomy. It is possible that the paleontologist, Tycho, and Kepler
should be unable to suspend the activation of the distinctive propensities and
to inhibit the production of the PI beliefs that each of them forms.
The theory-ladenness of perception has seemed epistemologically threat-
ening because it appears to block any chances of intersubjective agreement.
One aspect of the problem is whether certain kinds of changes in cognitive
state are impossible even for a normal scientific subject. The root worry is
that there might be rival systems of cognitive commitmentinvolving special
items in one or more of the long-term subsystemssuch that both would
accommodate the same perceptual stimuli in different ways so that no agree-
ment would ever be reached. Instead of providing an independent check on
cognitive commitments, perception would simply be a vehicle for persisting
in the prior state, and the only ways of moving scientists to consensus would
be through bribery, coercion, or some other form of conversion. Now the
issues involved in this arena are surely partly empirical: findings from per-
ceptual psychology can disclose whether people can come to adopt various
kinds of PI beliefs under various kinds of stimuli if they start with certain
kinds of prior commitments. However, as will become apparent in Chapter
7, it is important to have a general picture of the cognitive system, within
which alternative hypotheses about theory-dependence can be formulated.
11. Of course, it is by no means an easy matter to say exactly what this means. Since
my task here is to characterize the traditional notion of theory-independent observation
from the perspective of my general approach to cognition, I shall be generous and assume
that there is some way to make sense of the notion of proper functioning, perhaps in terms
of the coherence of the deliverances obtainable from all (or an overwhelming majority of)
the system's sensory receptors.
68 The Advancement of Science
4. Cognitive Variation
Are there particular forms that a cognitive system should take, particular
propensities that should be present within it, or particular goals that it seeks
to achieve? Legend conceives of scientific method as determinate. There are
no options for the good scientist.13 On one simple conception of scientific
method, it would be quite impossible for it to be permissible to believe either
of two incompatible propositions on the basis of some total evidence. If both
members of a pair of contraries were equally reasonable, then the proper
scientific stance would be agnosticism. Part of my rejection of traditional ideas
about method will consist in celebrating human cognitive variation.
Scientists differ in their relations to the phenomena: what is known to one
is not known to another, what is vivid for one is only a matter of "book
knowledge" for another, and so forth. These differences in relationships to
"the evidence" may prompt subjects to undergo quite different psychological
processes when making up their minds on controversial issues. Thus, in the
early twentieth century, naturalists who explored variation in wild populations
of butterflies and the geneticists studying mutations in Drosophila drew quite
different conclusions about the operation of selection, not simply because
they assented to different "evidential reports"indeed each group endorsed
the findings cited by the otherbut because their training disposed them to
12. As we shall see in Chapter 7, the problem of inaccessibility and the nature of the
underlying dynamic constraints also bear on the question of the discovery/justification
distinction.
13. Of course, options are allowed in subjectivist versions of Bayesian confirmation
theory. If it is supposed that a necessary and sufficient condition for rationality is proba-
bilistic coherence, then it is possible for rational scientists to be in extended disagreement
because of differences in prior probabilities. However, it is precisely this feature of sub-
jectivist Bayesian accounts that prompts the protest that they offer an inadequate repre-
sentation of scientific method.
The Microstructure of Scientific Change 69
feel the force of certain kinds of considerations and not that of others.14 At
an even more fundamental level, very basic cognitive strategies may vary.
Some people are cautious in drawing general conclusions, others far more
daring; some try to integrate their conclusions, others are content to pursue
independent lines of reasoning without thinking how they may be connected.
I want to contrast my ideas about cognitive variation with Legend's con-
ception that there is a unique set of principles that ought to govern people's
cognitive lives. This latter conception seems to me to generate the notion of
a properly functioning cognitive system, such that, given the same inputs from
experience, all properly functioning cognitive systems would undergo the
same sequence of states. We can articulate this idea within my framework as
follows: The cognitive system of a scientistany scientistcontains the same
set of goals and the same set of inferential propensities for working over input
from experience. In principle, we can suppose that all scientists receive the
same stimuli as inputs. Hence they should all produce the same output, storing
the same propositions in declarative memory, acquiring the same skills in
procedural memory, and so forthunless, of course, they are not properly
working scientific cognitive systems. There are two different ways in which
they may fail to work properly: they may fail to receive certain stimuli or
may receive misleading stimuli, so that they fail to acquire certain PI beliefs
or even acquire wrong beliefs (in which cases the scientist is ignorant or
misinformed), or their set of inferential propensities may lack correct pro-
pensities or contain wrong ones (in which case the scientist is logically deficient
or irrational).15 But, it is assumed, most scientific minds usually work well
and the failures are typically failures of ignorance or misinformation. The
philosophical task is to understand the proper working of the system of in-
ferential propensities.16
Our (primitive, but heuristically useful) cognitive geography exposes four
different types of cognitive variation. First, if scientists have access to different
14. In Giere's fruitful terminology, members of the two groups have different "cognitive
resources" (Giere 1988 213-214). Mayr's discussion of post-Darwinian evolutionary theory
in his (1982) is largely devoted to emphasizing the difference between the decision making
of naturalists and experimental geneticists. Perhaps as a corrective to other historical ac-
counts, he is concerned to praise the efforts of the naturalists. I would prefer to give a
more symmetric treatment.
15. We might add a further type of lapse to cover formation of the wrong kinds of
goals. Thus some scientists might put the interests of their careers or the advancement of
some social cause before their dedication to the epistemic goals of science. For the moment,
I shall ignore this sort of falling off since, on Legend's ideal of scientific behavior, those
who devote themselves to the wrong goals may not count as scientists at all. But, in Chapter
8, I shall begin to explore the social consequences of such "lapses."
16. On subjectivist Bayesian attempts to articulate Legend's ideas about method, there
is an important source of nonuniformity. Scientists differ in their prior probability distri-
butions, so that there is one blind and arbitrary difference in cognitive state that would
affect the ways in which inputs from experience are assimilated. It seems to me that this
approach generates the wrong kind of cognitive variation, assuming that people are uniform
in some relentless propensity for conditionalization but that there is enormous latitude in
their initial judgments of probability.
70 The Advancement of Science
bodies of information, then they may form quite different beliefs so that the
contents of declarative memory (and possibly of the set of long-term goals
and of procedural memory) will differ from scientist to scientist. Second,
propositions are likely to be stored in different ways in the declarative mem-
ories of different scientists, so that what is easily accessible for one scientist
is not readily produced in the problem solving/decision making of another.
Third, depending on the order and frequency of exposure to different types
of information, certain of the propensities may become more likely to be
activated than others. Two scientists may share the same propensities, in the
sense that the same repertoire is present in both, but different propensities may
be facilitated in the two, as a result of differences in intellectual development.
Finally, the scientists may differ in the contents of procedural memory, with
one having inferential propensities that are not found in the other.
As I have already remarked, Legend sees exposure to a truncated or
deceptive set of stimuli as producing ignorant or misinformed scientists. It is
assumed that the limitations could easily be made up. However, our reflections
on constraints should question this optimistic assessment. If no single scientific
mind can store all the propositions (say) that are relevant to the further
advancement of a field, then the differences among scientists are not accidental
but essential to continued growth: the development of the field would be
stunted if uniformity were imposed.17 However, such differences may affect
the ways in which further stimuli are accommodated, for, as we saw, there
may be dynamic constraints on cognitive processing. But if Legend's treatment
of the universal accessibility of information is too facile, its response to the
fourth type of cognitive variation, the possibility of different propensities in
the procedural memories of different subjects, seems to me to be too harsh.
As I shall explain, it is highly likely that variation in these propensities plays
a valuable role in the growth of science.
Before I take up this topic, let us look briefly at the types of cognitive
variation that have traditionally been ignored. External aids, especially text-
books, handbooks, and review articles, contribute to the homogenization of
scientific experience, helping to foster the illusion that the same information
is "in principle accessible" to all the workers in a scientific field. That this is
an illusion becomes apparent when we reflect on situations of problem solving
or decision making. For each scientist, there are particular items in declarative
memory that are readily activated, others that are activated only with diffi-
culty, particular pieces of information that can be assembled with ease from
books and colleagues, others that are very hard for the scientist to track down.
In thinking about evolutionary problems and scenarios, Darwin was surely
guided by his direct experiences with South American mammals, Dobzhansky
by his wild Drosophila populations, Gould by his fieldwork on the snail
Cerion, E. O. Wilson by his deep knowledge of ants. Other connections are
less direct: Darwin had secondary, but still privileged, access to information
about pigeons and arctic flora, Gould to facts about trilobites, Wilson to
17. This theme has been taken up and elaborated by Miriam Solomon (1992).
The Microstructure of Scientific Change 71
knowledge about termites. At the other extreme, there are items that are
now virtually inaccessible: collections of fossil shells that have remained for
years in unopened drawers, beautifully prepared illustrations in nineteenth
century works of natural history that no one has read in decades. In general,
we can think of the declarative memories of scientists and their relations to
others and to records (books, reviews, samples, etc.) as determining a system
of nested spheres. The items in the interior spheres are easily recovered and
play a prominent role in the scientist's reasoning. Those at the periphery are
effectively beyond the scientist's ken. As I shall argue much later in this book,
it is likely that efforts to impose a uniform ordering across a scientific field
would severely hamper that discipline's further development.
Similar points can be made about differences in the activation of pro-
pensities. Co-workers in a scientific discipline often recognize one another as
having distinctive cognitive styles. Faced with the same problem, different
scientists will pose it and reason about it in different ways.
Once one has raised the possibility that there may be different propensities
for activating modes of problem solving, representation, and inference that
would be widely regarded as respectable, few people are likely to deny that
such variation can be healthy in the growth of science. I now want to consider
the far more radical claim that there may be significant differences in the
fundamental inferential propensities stored in procedural memory, and that,
contrary to the ideal of methodological uniformity assumed by Legend, this
type of diversity can also be beneficial.
Any system that needs to make ampliative inferences (and human brains
are among the many systems that are required to do so) has to strike a
compromise between waiting too long and jumping to conclusions too quickly.
It is quite possible that the tradeoff is made in a stable way for adult scientists,
so that, for each scientist, there is a definite propensity (or spectrum of
propensities) for that scientist to decide that she has enough instances, and
that this propensity (or spectrum) is variable across the scientific population.18
To put the point simply, some scientists may be jumpers and some inclined
to hang backHuxley and Owen serve us as possible exemplars.
There is no optimal height for a human male, one that will maximize
reproductive success across all human environments. So too the variation in
inferential success may vary, for jumpers and stick-in-the-muds, depending
on the kinds of problem situations in which those propensities are called into
play. If so, there will be no sense to the ideal of a single method, a unique
description of the "properly functioning scientific mind." However, just as
the recognition of biological variation does not prevent us from recognizing
that certain alleles (and their bearers) are grossly nonfunctional, so too our
18. Howard Margolis (1987) offers some illuminating ideas about this topic, suggesting
that there has to be a balance among four "subprocesses" he takes to be basic: jumping,
checking, priming, and inhibiting. During ontogeny brains have to become configured in
a dynamic system in which these four subprocesses interact. As I shall suggest later, there
may be no solution that is optimal across all experiential contexts, and the compromises
may be struck in quite differentadequateways in different instances.
72 The Advancement of Science
appreciation of cognitive variation should not blind us to the fact that certain
kinds of cognitive systems are (or would be) incompetent.19
I conclude that there are four important sources of cognitive variation and
that in none of the cases should we take it for granted that there is a uniquely
right way for the scientific cognitive system to be set up. Indeed, as I shall argue
at some length in the final chapter, cognitive variation contributes in several
important ways to the community project of investigating the world. Diversity
of the kinds I have discussed serves the community as a hedge against becoming
stuckmuch as variation in natural populations can enable a lineage to survive
events in which monomorphic populations would have gone extinct.
5. Goals
Legend takes it for granted that scientific behavior can and should be directed
by the desire to attain cognitive (and only cognitive) goals. But there are
important ambiguities here. Let us start by recognizing that scientists may
have the goal of advancing the cognitive well-being of the large intellectual
community to which they belongthe totality of scientists, past, present, and
future. They may also have the goal of improving their own cognitive state.
There is good behavioral evidence for attributing both goals. Biologists work-
ing on a particular group of organisms, Drosophila geneticists for example,
sometimes conceive themselves as contributing to a community project that,
as a whole, will promote our understanding of flies, inheritance, and evolution.
But, on other occasions, they confess to more personal aims. They just want
to understand (they say) how much polymorphism there is at such-and-such
a locus in such-and-such a population.
This last declaration is itself ambiguous. Would the imagined geneticist
be satisfied if someone else (perhaps someone drawing on the geneticist's
own partial results) measured the extent of polymorphism and obligingly
communicated the information? Probably not. A more precise formulation
of the geneticist's aim is to say that he hopes to find out how much poly-
morphism there is at the locus. The geneticist's goal is not simply to reach a
particular cognitive state but to arrive at it in a particular wayeither as the
result of his own efforts or through the work of a team to which he belongs.
Epistemic and nonepistemic considerations easily blur. What is it exactly
that the geneticist wants besides knowledge that there are precisely so many
alleles at the locus? The pure joy of discovery? The satisfaction of being the
first to know? Recognition for being the first to know? The increased authority
or increased resources that may accrue to the discoverer? All of these, per-
haps. Part of the story is surely that answering outstanding scientific questions
19. I shall elaborate these points at much greater length in Chapter 6. Throughout this
section, I have been much influenced by conversations with Stephen Stich and by his (1990).
Stich would, I think, go much further in celebrating cognitive variation. Much of the
difference can be traced to my attempt to defend the idea of broadly shared cognitive goals,
and to use the attainment of those goals as a test of the adequacy of cognitive systems.
The Microstructure of Scientific Change 73
6. Individual Practices
The most detailed picture of the life of a scientist would chart the moment-
by-moment changes in her cognitive state, the ways in which items from long-
term subsystems are activated and stored. At the opposite extreme, my ideal-
ized life cycle allows for a view of scientific change that abstracts from indi-
vidual scientists altogether, concentrating on the sequence of consensus
practices. But to understand why consensus practice changes in the way that
it does we need to consider the actions and attitudes of members of the
pertinent scientific community without taking into account all the fluctuations
that occur. It is therefore useful to have an intermediate level of description,
one that can be directly linked to the consensus practice of the community
and employed to understand changes in consensus practice but one that can
also be related to the psychological lives of individuals.
Generalizing from my account of the career of Darwinism, I shall take a
scientist's practice to be a multidimensional entity whose components are the
following:
1. The language that the scientist uses in his professional work.
2. The questions that he identifies as the significant problems of the field.
3. The statements (pictures, diagrams) he accepts about the subject matter
of the field.
4. The set of patterns (or schemata) that underlie those texts that the
scientist would count as explanatory.
5. The standard examples of credible informants plus the criteria of cred-
ibility that the scientist uses in appraising the contributions of potential
sources of information relevant to the subject matter of the field.
6. The paradigms of experimentation and observation, together with the
instruments and tools which the scientist takes to be reliable, as well
as his criteria for experimentation, observation, and reliability of
instruments.
7. Exemplars of good and faulty scientific reasoning, coupled with the
criteria for assessing proposed statements (the scientist's "meth-
odology").
In each of the last three components, there are two different levels to the sci-
entist's practicea commitment to cases (typically reflected in behavior) and
an embryonic theory about why the behavior is correct. If pressed about his
provocative study (Sulloway forthcoming), Frank Sulloway reviews the attitudes of scientists
at times of large ("revolutionary") change, arguing that scientists' responses are shaped by
"psychologically formative influences operating within the family." I am extremely grateful
to Sulloway for sharing his work with me.
The Microstructure of Scientific Change 75
7. Scientific Language
Scientists formulate their ideas in extensions of natural languages. Correlating
the scientific discourses carried out in two different extended natural languages
is frequently very easy: the technical expressions of one language (neologisms
or old terms stretched to new uses) map smoothly onto the technical terms of
the other. Yet, ever since Kuhn (1962/70), philosophers have been challenged
by the suggestion that the languages employed by scientists who are separated
by a major upheaval in their field are not intertranslatable. How should we
think about scientific language so as to become clear about scientific communi-
cation and its limits and so as to recognize the dynamics of conceptual change?
The crucial issue concerns the semantics of the language. Syntax is rela-
tively unproblematic, for we may regard the grammar of scientific language as
that of the natural language in which it is embedded. Since Frege's seminal dis-
cussion (Frege 1892), it has been clear that linguistic expressions typically have
two semantic functions. One of these is to refer (or to purport to refer) to enti-
ties: objects, sets of objects, relations, and so forth.22 However, Frege argued
that reference could not exhaust the semantic functioning of language, for, in
many cases, replacing a constituent of a complex expression by a co-
referential expression affects the "content" of the whole. Preservation of ref-
erence does not suffice for preservation of meaning. To use Frege's famous ex-
ample, the expressions "the morning star" and "the evening star" both refer to
the planet Venus, but, whereas the sentence "The morning star is the same as
the evening star" is informative (and could be surprising to a hitherto unin-
formed astronomical observer), the sentence "The morning star is the morning
star" is a trivial identity that will be accepted by the astronomical tyro.
Frege claimed that such examples revealed the existence of intensional
entities, senses, that are expressed by linguistic expressions. The language
user who understands an expression grasps its sense.23 Sense also determines
reference: two expressions with the same sense must share the same reference.
Recent discussions in the philosophy of language have made it evident that
22. The standard semantics for the formal languages that logicians study, a semantics
that descends from Frege but that modifies his views in important ways, takes names to
refer to objects (people, organisms, physical things, and perhaps abstract entities: sets,
numbers, concepts, ideas, and so forth). One-place predicates (e.g., "is wise," "is round,"
"has a mass of 30 grams") refer to sets of thingsin each case the set of things that satisfy
the predicate in question. Two-place predicates (such as "is larger than") refer to sets of
ordered pairs of objects: an ordered pair (x, y) belongs to the referent of "is larger than"
just in case x is larger than y. And so it goes.
23. The most extensive exposition of Frege's views about language in general and his
ideas about sense in particular is provided by Michael Dummett in his (1973).
76 The Advancement of Science
the conjunction of the two ideas is problematic (Kripke 1972, Donnellan
1974, Putnam 1973). Part of the problem stems from the fact that competent
users of expressions are often in no position to identify criteria that the
referents of the expressions must meet: people can talk about stoats and
weasels, or elms and beeches (Putnam's example), without knowing what
differentiates a stoat from a weasel or an elm from a beech. Apparently, it
is enough for us to refer that there be experts, on whom we rely, who can
make the distinctions. Another facet of the problem is that, even when a
language user has a relatively rich set of beliefs about the referent, what she
grasps ("what is in the speaker's head" in Putnam's phrase) may not suffice
to isolate the referent under all possible circumstances. The slack may be
taken up by the way the world is and by the relationship between the person
and the world. Thus, to recall a thought experiment of Putnam's, a person
on Earth and her molecule-for-molecule replica on twin Earth (her Doppel-
ganger) use the term 'water' to refer to the colorless, odorless liquid that they
drink and bathe in. Each of them has all the everyday beliefs about water
that nonscientists standardly enjoy. However, while on Earth the liquid is
H2O, on twin Earth it is a different substance XYZ. Despite the identity of
their conceptions, the earthling and her Doppelganger refer differently, the
one to H2O, the other to XYZ. "What is in the speaker's head" does not
therefore determine reference. I shall articulate my approach to scientific
language by building on the recent insights about reference.24
Imagine that a speaker produces a token of a term, "the morning star"
say, and thereby refers to an object. Somehow a connection is made between
the noises that the speaker produces and a part of nature. I shall call what
makes it the case that that token refers to that object the mode of reference
of the token.25 For any case in which someone uses an expression successfully
to refer, we can ask what makes it the case that the token refers to the entity
it doeswhat is the mode of reference of the token?
The explanation may take the following form: the present speaker intends
to continue her own prior references using tokens of the type; her own ref-
erential practice may have been begun by acquiring the term from others to
whose usage she intended to conform; those others may have adopted the term
24. The recent literature on reference, sense, and the important problems raised by
Kripke, Putnam, and Donnellan is enormous. Major contributions have been made by
Gareth Evans (1975), Michael Devitt (1981), and others. I shall attempt to remain neutral
on many of the controversial issues, since the problems that arise in dealing with conceptual
change in science seem to be largely independent of questions that philosophers of language
rightly take very seriously.
25. "Mode of reference" is a Fregean term, and Frege sometimes thinks of the sense
of an expression as its mode of reference (1892). However, as several writers (most notably
Hartry Field (1977) and Tyler Burge (1976)) have pointed out, the notion of sense is required
to play a number of roles in Frege's system (the sense of an expression is to be that which
is grasped by someone who understands the expression and also to be that which is the
reference of tokens of the expression that occur within contexts of indirect discourse).
Problems for Frege's approach (including Putnam's twin earth puzzle) arise because no
single thing can play all the roles Frege attributes to senses.
The Microstructure of Scientific Change 77
from yet others; and so it goes until we reach a first user (or first users) who in-
troduced the term by uttering it in the presence of an object with a present in-
tention to give the name to that object. In this case, let us say that the mode of
reference of the token that is presently produced by our imagined speaker is the
complex causal chain that stands behind her current vocalization. That the to-
ken has the reference it does is partly determined by the speaker's intentions (if
she did not intend to continue her own earlier usage and the usage of others,
then the history of prior usage would be quite irrelevant to the reference of her
token). It is also partly determined by events, states, and processes that are in-
dependent of her, to wit the transmission of the term and the connections be-
tween the first user(s) and the ostended object.26
Different tokens of the same type may be associated with different modes of
reference. Imagine our speaker on a later occasion, scanning the early morning
sky with a friend: "Look!," she exclaims, "the morning star is very low above
the horizon." On this occasion, her dominant intention may not be to agree
with her own prior usage or the usage of her community, but to single out a par-
ticular object in her line of vision. In effect, she is assuming the role of the first
user(s) and refixing the reference of the term. She believes, of course, that the
object singled out is the very same object that she has previously identified
when taking over the usage of others, but, if she were to discover that she were
wrong, her dominant intention is to talk about that thing, the heavenly body
over there near the horizon. On yet other occasions she may intend to pick out
an object by description, identifying the morning star as "the heavenly body
that appears close to the sun in the morning" or "the second planet from the
sun." Her intention may be revealed in her linguistic behavior, as when she rea-
sons as if she were excluding, by linguistic fiat, the possibility that the morning
star could fail to satisfy a certain description.27
Modes of reference thus fall into three types. A token's mode of reference
is of the descriptive type when the speaker has a dominant present intention
to pick out something that satisfies a particular description and the referent
of the token is whatever satisfies the description. The baptismal type is ex-
emplified when the speaker has a dominant present intention to pick out a
particular present object (or a set of objects, one member of which is present).
Finally, the conformist type covers those (many) instances in which the speaker
intends that her usage be parasitic on those of her fellows (or her own earlier
self), and, in this case, the reference of her token is determined through a
26. Those familiar with the seminal writings of Kripke and Putnam (especially Kripke
1972, Putnam 1973) will recognize that I am greatly compressing points that they elaborate
in rich detail. Both writers see a role for ostension in the initial determination of reference,
Kripke provides some excellent examples of how the reference of names can be fixed
through causal chains extending back into history through the usages of earlier speakers,
and Putnam emphasizes our dependence on experts ("the division of linguistic labor").
27. What is in the speaker's head thus determines something. As I shall suggest later,
there are frequently several available ways of fixing the reference of a token of a given
type, and the speaker's cognitive statespecifically an activated intentionpicks out which
of these fixes the reference of her current token.
78 The Advancement of Science
long causal chain that leads back to an initial usage, a usage in which a token
produced by a first user has its reference fixed either in the descriptive or in
the baptismal mode.
Within a particular scientific community a particular term type is associated
with a compendium of modes of reference that cover the variety of ways in
which, for members of that community, the reference of tokens of that type
may be determined. There may be a number of shared "first usages" that can
serve as termini for conformist modes of reference. There may also be several
descriptions (presumed to be equivalent) that members may employ to fix the
reference of a token of the type, and, if inferences depend on the satisfaction of
such a description, then, if this is made clear, the inference will not be chal-
lenged. Finally, although baptismal modes of reference may be idiosyncratic,
in the sense that speakers may baptize the same objects on quite different oc-
casions or may baptize different members of the same set or samples of the
same stuff, there will be significant common properties of the baptismal events:
"the morning star" will always be employed to dub an object visible in the sky,
"water" will always be employed to dub a sample of a liquid, and so forth.
I shall call the compendium of modes of reference for a term (type) the ref-
erence potential of that term. An important part of my story about conceptual
change in science (see Chapter 4) will be that reference potentials are typically
heterogeneous: that the linguistic community to which a scientist belongs
allows a number of distinct ways of fixing the reference of tokens of terms.
What, if anything, can be said about Frege's original problem? How do
we explain the difference in cognitive content between "The morning star is
the same as the evening star" and "The morning star is the morning star"?
Instead of positing new objects, senses, and a mysterious process by which
the speaker "grasps" them, let us develop our picture of cognition by intro-
ducing the notion of reference potential. For ease of exposition, I shall sup-
pose that acquiring the reference potential of a term consists in incorporating
a set of propensities into procedural memory. Imagine, then, that a speaker
has acquired reference potentials for "the morning star" and "the evening
star," and that these are not the same. If the speaker considers the sentence
"The morning star is the same as the evening star," it is likely that the tokens
of "the morning star" and "the evening star" will have their references fixed
by differing modes of reference. If this is the case, then, I shall assume, the
decision to determine whether or not the sentence is true activates a search
of declarative memory. By contrast, it is probable that both tokens of "the
morning star" in "The morning star is the same as the morning star" will
have their reference fixed in the same way. In this case, consideration of the
truth value of the sentence activates (at least in those speakers who have
acquired the use of "is the same as" in English) a propensity to assent to it.
Thus while the one sentence is assented to if it can be retrieved from declar-
ative memory (and then on the basis of the process of retrieval) the other is
judged correct by English speakers on the basis of a quite different process.28
28. I owe to the referee the interesting point that I can know the truth of "a = a"
The Microstructure of Scientific Change 79
even in instances in which I have no reference potential for V at all. I would suggest that
this is because I can recognize that whatever the modes of reference of the two tokens of 'a'
might be they would fix reference to the same object. Again, there would be an important
cognitive difference with statements of form "a = b."
29. My account shifts the focus from discussion of senses (abstract objects to which we
stand in mysterious relations) to consideration of grasp of senses (conceived of as psycho-
logical processes that can be assigned places within our cognitive geography). Once we
have linked these psychological processes to the ways in which reference is fixed we have
a natural way of treating Frege's puzzle.
30. The problem is posed forcefully by John Dupre in his (1981), and I attempted to
respond to it in my (1982), on which the discussion of the next paragraph is parasitic.
Michael Devitt and Kim Sterelny have some illuminating things to say about this issue
which they call "the qua problem"in their (1987).
80 The Advancement of Science
would discriminate certain things as relevantly similar and others as not. There
is no reason to think that these discriminatory dispositions as I shall call them
are completely determinate. But, I shall imagine, if there is a unique natural
kind that includes the object ostended and that conforms to the discriminatory
dispositions, then the kind is the referent of "tiger."
Introduction of the notion of natural kind brings us to a third feature of
my account that stands in need of explicit discussion. Some, but not all, of
the predicates of the language are marked as referring (under a certain mode
of reference) to natural kinds. Such predicates are associated with propensities
for use in explanation and in inductive generalization. Learning a language,
according to this conception, involves acquiring propensities for forming cer-
tain types of generalizations. Alternatively, the language embodies a view of
where the divisions in nature are, and to learn the language is to acquire a
propensity to see those divisions as natural. I shall have more to say about
these issues later (see chapters 4, 5, and 7).
scientist honors the impersonal goal of devoting his labor to the advancement
of the field, he commits himself to working on these questions and to assessing
their importance in the manner delineated in the partial ordering. However,
the scientist has concerns of his own, matters that particularly interest him
or to which he believes that his talents are especially suited. Such concerns
generate a personal assessment, which can also be regarded as a partial or-
dering of a set of questions, the questions that the scientist views as significant
for him to try to answer. We should not expect that these assessments match
either in the membership of the sets or in terms of the orderings.
The third component of the practice of a scientist at a time is the set of
statements about the subject matter of the field to which the scientist would
assent at that time. This component recalls the traditional philosophical rep-
resentation of the state of a science at a time. But there are some significant
differences. First, I shall not insist that this set contain any general (lawlike)
statements. As our study of evolutionary biology revealed, there are large
subfields within biology that appear to lack any distinctive general principles
of their own.32 The generality that they achieve seems to be attained through
the repeated use of explanatory patterns. Second, I shall not assume that the
set is deductively closedindeed, it may well be finite. A straightforward way
to resist the demand for deductive closure is to recognize that the set of
statements that scientists accept about the subject matter of their field might
be subtly inconsistent (certainly, if one includes in the set reflective judgments
prompted by epistemic modesty, such as the judgment that one of one's beliefs
about nature is false, there will be an inconsistency). If the set is inconsistent
then every statement of the language belongs to its set of consequences, so
that if it were deductively closed the scientist would be committed to accepting
all statements of the language.33
Yet, even though the set of accepted statements does not have the structure
of an axiomatic deductive system, it nevertheless has an important structure.
The scientist is disposed to take some statements he accepts as providing his
reasons for accepting others: the psychological processes through which he
would currently sustain his beliefs in the latter statements involve the acti-
vation of inferential propensities that start with beliefs in the former state-
ments. The justificatory structure latent in the scientist's psychological life at
any time may be radically incomplete. With respect to many statements the
scientist may be able to say little more than "I remember it but I don't know
where I learned it" or "I think I got that from an article" or "I learned that
in graduate school" or "That was demonstrated by Crumkopf and Blither-
stein." Sometimes a statement may be defended with the claim that some
entity, state, or process was observed. Frequently, however, only the summary
of the data, the product of much analysis, will be retained. The scientist
supports her statement by declaring, "I (we) did an experiment that showed
32. I have also argued this point for the case of genetics in my (1984) and (1989).
33. For elaboration of this argument and some morals about requirements on human
rationality, see (Stich 1990).
82 The Advancement of Science
that," but, even if the circumstances of the experiment can be identified and
the lab notebook is extant, the scientist will still have to rely on the authority
of his own earlier recording of measurements and his own prior analysis of
the data (unless, of course, he chooses to revise the latter).34
9. Explanatory Schemata
A crucial part of a scientist's practice consists in her commitment to ways of
explaining the phenomena. Faced with explanation-seeking questions, the
scientist is disposed to produce texts instantiating particular patterns. These
patterns, or schemata, although they are not likely to be formulated by sci-
entists themselves (at least not in the style of my reconstruction of Darwinian
patterns), are implicit in scientific practice, and I would expect that practi-
tioners could recognize them as underlying their own explanations.
What is a pattern? Let me give a general account of the conception I
introduced with examples in Chapter 2. A schematic sentence is an expression
obtained by replacing some, but not necessarily all, the nonlogical expressions
occurring in a sentence with dummy letters. Thus, starting with the sentence
"Organisms homozygous for the sickling allele develop sickle-cell anemia,"
we can generate a number of schematic sentences: for example, "Organisms
homozygous for A develop P" and "For all x, if x is O and A then x is P"
(the latter being the kind of pattern of interest to logicians, in which all the
nonlogical vocabulary gives way to dummy letters). A set of filling instructions
for a schematic sentence is a set of directions for replacing the dummy letters
of the schematic sentence, such that, for each dummy letter, there is a di-
rection that tells us how it should be replaced. For the schematic sentence
"Organisms homozygous for A develop P," the filling instructions might
specify that A be replaced by the name of an allele and P by the name of a
phenotypic trait. A schematic argument is a sequence of schematic sentences.
A classification for a schematic argument is a set of statements describing the
inferential characteristics of the schematic argument: it tells us which terms
of the sequence are to be regarded as premises, which are inferred from
which, what rules of inference are used, and so forth. Finally, a general
argument pattern is a triple consisting of a schematic argument, a set of sets
34. Large parts of the set of accepted statements in a scientist's practice are there
through doxastic inertia. If the scientist now became committed to some Cartesian enterprise
of overhauling his system of beliefs and resolved to throw out everything he could not
justify as of now, then he would be left with a severely mutilated corpus. Gilbert Harman
(1987) argues persuasively that there is an important difference between realizing that your
old reasons are poor and realizing that you can no longer formulate what strike you as
convincing reasons: in the former, but not the latter, instance there are grounds for un-
dertaking belief revision. This is a helpful contrast, but it is only the beginning of the
investigation of the desirability of doxastic inertia. As one might expect, the presence of
some self-critical people within a scientific community, people who are prepared to give
more weight to the traditional ideal of abandoning what is currently unjustified, can some-
times prove valuable for the advancement of the community project.
The Microstructure of Scientific Change 83
of filling instructions, one for each term of the schematic argument, and a
classification for the schematic argument.
A particular derivation, the sequence of sentences and formulae found in a
scientific work for example, instantiates a general argument pattern just in case
(i) the derivation has the same number of terms as the schematic argument of
the general argument pattern, (ii) each sentence or formula in the derivation
can be obtained from the corresponding schematic sentence in accordance with
the filling instructions for that schematic sentence, (iii) the terms of the deriva-
tion have the properties assigned by the classification to corresponding mem-
bers of the schematic argument. The Darwinian patterns of the last chapter are
general argument patterns in the sense introduced here, and the explanations
produced by Darwin, Huxley, Fisher, Wright, Dobzhansky, Simpson, and
other luminaries in the Darwinian tradition instantiate these patterns in the
fashion just defined. Consider one of the simplest of my examples:
COMMON DESCENT
Question: Why do the members of G, G' share P?
Answer:
(1) G, G' are descended from a common ancestor G0.
(2) G0 members had P.
(3) P is heritable.
(4) No factors intervened to modify P along the G0-G, G0-G'
sequences.
Therefore (5) Members of G and G' have P.
This consists of five schematic sentences. We could equip it with Filling In-
structions by requiring that G, G', G0 be replaced by names of groups of
organisms, and that P be replaced by the name of a trait of organisms.
Similarly, the classification would declare that (l)-(4) are premises and that
(5) is deduced from them.
Derivations may be similar in terms of either their logical structure or the
nonlogical vocabulary they employ at corresponding places. The notion of a
general argument pattern allows us to express the idea that derivations similar
in either of these ways share a common pattern. However, similarity is a matter
of degree. At one extreme, a derivation is maximally similar to itself and to it-
self alone; at the other, any pair of arguments can be viewed as sharing a com-
mon pattern. To capture the notion that one pair of arguments is more similar
than another pair, we need to recognize the fact that general argument patterns
can demand more or less of their instantiations. If a pattern sets conditions on
instantiations that are more difficult to satisfy than those set by another pat-
tern, then I shall say that the former pattern is more stringent than the latter.
The stringency of an argument pattern is determined in part by the clas-
sification, which identifies a logical structure that instantiations must exhibit,
and in part by the nature of the schematic sentences and the filling instructions,
84 The Advancement of Science
1974). Interestingly, the central point was never expressed better than by the onetime
logical positivist Otto Neurath in the passage that Quine uses at the beginning of his (1960):
"We are like sailors who must always rebuild our ship on the open sea without ever
anchoring in dry dock."
86 The Advancement of Science
scientists face. For some exploration of this in the particular case of mathematics, see
chapters 8-10 of (Kitcher 1983a).
39. There are obvious links between my notion of consensus practice and the Kuhnian
concept of paradigm. Like Kuhn I take what is shared among scientists to be far more than
the acceptance of a set of statements. However, Kuhnian paradigms are intended both to
offer a richer description of the state of a science at a time and to divide the history of a
scientific discipline into segments, such that there are significant epistemological differences
between the course of science within a segment and the intersegment transitions. My
approach bears no commitment to this second function. See (Kitcher 1983a chapter 7).
88 The Advancement of Science
1. A Problem Reposed
In conceiving of science as progressive we envisage it as a sequence of con-
sensus practices that get better and better with time. Improvement need not
be constant. Like the fortunes of a firm, the qualities of consensus practices
may fluctuate. But, if a scientific field is making progress, there should be a
general upward trend, comparable to the generally rising graph of profits that
the firm's president proudly displays at board meetings.
The challenge is, of course, to say what "better and better" means in the
scientific case, what corresponds to the accumulating dollars, yen, marks, etc.
It is important to be clear from the start that the first task is to pinpoint those
relations that should obtain among the members of a sequence of consensus
practices if that sequence is to be progressive. To ask how we know that a
field of science is making progress is to pose a separate question. There is an
important connection between the questions: a definition of progress would
be of little value if there were good reasons for thinking that it is impossible
for us ever to know that we are making progress in the specified sense. Despite
the connection, we should not confuse the issues, for example by lambasting
a proposal for defining the notion of progress with the complaint that it has
not yet been explained how we know that we are making progress. In my
treatment, the definition will come first (this chapter). A response to various
kinds of worries about concepts that I use will be presented later (Chap-
ter 5).
Because one species of concern is so prevalent, it is worth forestalling
related criticisms before we begin. I shall sometimes make free use of the
concept of truth, explicating certain types of scientific progress as involving
the attainment of truth. Moreover, I shall make judgments about the record
of various scientists, past and present, in attaining truth, judgments based on
the understanding of nature furnished by our current science. If the uses I
make of the concept of truth are incoherent, then my account of progress
will be wrong. Similarly, if the judgments about past and present attainment
of truth that I make are incorrect or unjustified, then my errors will infect
either my analysis of scientific progress or my claims about the ways in which
science has made progress. Thus there are important objections that I shall
90
Varieties of Progress 91
rowness of the set of impersonal epistemic goals.2 Although the task of pro-
viding an account of practical progressiveness is an important one, it should
follow the more tractable project of understanding cognitive progress.
progress. But, in my judgment, truth is not the important part of the story.
Truth is very easy to get.4 Careful observation and judicious reporting will
enable you to expand the number of truths you believe. Once you have some
truths, simple logical, mathematical, and statistical exercises will enable you
to acquire lots more. Tacking truths together is something any hack can do
(see Goldman 1986 123, Levi 1982 47, and, for the locus classicus of the point,
Popper 1959 chapter 5). The trouble is that most of the truths that can be
acquired in these ways are boring. Nobody is interested in the minutiae of
the shapes and colors of the objects in your vicinity, the temperature fluc-
tuations in your microenvironment, the infinite number of disjunctions you
can generate with your favorite true statement as one disjunct, or the prob-
abilities of the events in the many chance setups you can contrive, with objects
in your vicinity. What we want is significant truth. Perhaps, as I shall suggest
later (Section 8), what we want is significance and not truth.
Scientists in the tradition that extends beyond the seventeenth century to
the ancient Greeks have been moved by the impersonal epistemic aim of
fathoming the structure of the world. In less aggressively realist language,
what they have wanted to do (as a community) is to organize our experience
of nature. Seventeenth century thinkers such as Boyle and Bacon attributed
special significance to this project because, seeing nature as God's Creation,
to expose the structure of nature was to recognize the divine intentions. That
further link, useful, possibly even essential, though it may have been in
showing how science contributes to our overall conception of what is valuable,
can be discarded if we merely pursue the internal project of understanding
our impersonal epistemic goals. Our task is to make more precise the murky
notion of uncovering the structure of nature (or of organizing experience),
of discovering, insofar as we are able, how the world works.5
Historians have been struck by the apparent variability of the goals pro-
fessed by different scientists, working in different communities at different
times. But I want to start from the opposite end. It is hardly surprising that
4. Conceptual Progress
Conceptual progress is made when we adjust the boundaries of our categories
to conform to kinds and when we are able to provide more adequate speci-
96 The Advancement of Science
fications of our referents.6 Striking examples come from the history of all
sciences: 'planet,' 'electrical attraction,' 'molecule,' 'acid,' 'gene,' 'homology,'
'Down's syndrome' are all terms for which faulty modes of reference have
been improved. Consider, for example, pre-Copernican usages of 'planet.'
What set of objects was singled out? One answer is that, for the pre-
Copernican, as for us, planets are Venus, Mars, and things like thatin which
case the planets would include the Earth. Another is to declare that the planets
are specified as those heavenly bodies that are sometimes observed to give
rise to erratic motions against the sphere of fixed stars (the "wanderers")
in which case the planets would be Mercury, Venus, Mars, Jupiter, and
Saturn. Or, allowing the specification to take account of possible perceptual
limitations, we might declare that the planets comprise not only those ob-
served to "wander" but those we would observe "wandering" if we had keener
senses, that is, all the things we would count as planets of the solar system
except the Earth. If we adopt the first approach to the reference of 'planet'
then pre-Copernicans succeed in marking out a natural division, but they fail
disastrously when they come to say what they are talking about. On either
of the second approaches, they are fully aware of what they are referring to
but the set of objects they single out is not a natural category. Either way,
post-Copernicans do better. And so it goes, not only for this example but for
the other terms on my brief list and for many more besides.
As I shall explain later, the answers of the last paragraph are not entirely
adequate, but they enable us to appreciate the possibility of understanding
conceptual change and conceptual progress in terms of shifts in mode of
reference. The thesis that I shall defend in this section is that the conceptual
shifts in science that have caused most attention (and that are supposed to
be troublesome) can be understood, and understood as progressive, by rec-
ognizing them as involving improvements in the reference potentials of key
terms. Recall from Section 7 of the last chapter that the reference potential
of a term for a community is a compendium of the ways in which the references
6. Here I shall take over, uncritically for the time being, Plato's famous metaphor of
"carving nature at the joints." So part of the task of fathoming the structure of nature will
be seen as an exercise in anatomy, exposing the lineaments of the objective divisions of
nature and showing which things belong together. A strong realism about the structuring
of nature provides a simple way of developing the fundamental epistemic aim selected in
the last section, but it brings well-known epistemological difficulties in its train. I shall
suggest later (Chapter 5 Section 8) that there are other metaphysical options, which seek
to explicate the notion of natural kind by focusing on the groups to which reference is
made in laws or in explanatory schemata, which can be linked to the project of fathoming
the structure of nature, and which can avoid some of the epistemological difficulties. As I
have suggested elsewhere (Kitcher 1986, 1989) I believe that this latter approach to the
metaphysical issues has important advantages. However, the treatment of many questions
about progress can profitably be undertaken by leaving the notions of natural kind and
objective dependency (see Section 5) temporarily unanalyzed, using them to explicate
conceptions of progress, and only returning to metaphysics later. So this chapter will remain
neutral between the options of strong realism and the type of weakened (Kantian) realism
that I have previously espoused.
Varieties of Progress 97
of tokens of that term can be fixed for members of the community. I shall
illustrate how reference potentials may shift by analyzing an apparently prob-
lematic case, and then offer a general view about improvement of reference
potentials.
Kuhn and Feyerabend achieved the important insight that different com-
munities of scientists, working in the same field, may organize the aspects of
nature that concern them in different ways (Kuhn 1962/70, 1983; Feyerabend
1963a, 1965, 1970). Different conceptualizations make differences to obser-
vation, inference, and explanation (see Chapter 7).7 But, most famously,
Kuhn and Feyerabend have suggested that the phenomenon of "conceptual
incommensurability" that they identify engenders problems of communication
and of reporting of evidence. I shall begin by reviewing one of Kuhn's favorite
problematic cases, with a view both to showing that there is indeed an in-
teresting conceptual shift and to revealing that it does not have the dire
consequences for communication with which it is sometimes credited.
There is a very simple story about the chemical revolution of the eighteenth
century. Priestley, so the story goes, employed a language containing terms
'phlogiston,' 'principle'that fail to refer. Lavoisier used a language con-
taining expressions'oxygen,' 'element'that refer to kinds that Priestley
could not have identified. So there is a conceptual advance between Priestley
and Lavoisier, one that involves the replacement of expressions that fail to
refer by genuinely referring expressions and the introduction of terms that
single out kinds for the first time.
Kuhn saw that this story is extremely implausible. If Priestley's language
was so inadequate, then it is hard to understand how his chemistry could ever
have appeared successful and how he could have contributed to the devel-
opment of Lavoisier's own ideas. Yet Priestley was a successful chemist, one
who appeared to his contemporaries to be extending the range and power of
the phlogiston theory, and he was responsible for the experiment in which
he (and, following his lead, Lavoisier) isolated a new gas, the gas Lavoisier
called "oxygen." At times Kuhn seems to want to solve the problem of
Priestley's success by declaring that the terms we take to be nonreferring
('phlogiston' and so forth) really did refer, picking out constituents of the
"different world" inhabited by Priestley and his fellow phlogistonians. Di-
vided by the chemical revolution, Priestley and Lavoisier inhabit different
worlds (or, to put it another way, their theories have different ontologies).
Unfortunately, this strategy of accounting for Priestley's apparent success will
only accommodate part of the historical evidence. As we insist on the "many
worlds" approach, the revolutionary divide between Priestley and Lavoisier
looks ever more difficult to bridge, and communication between the two men
7. Thus, in the writings of Kuhn (1962/1970), Feyerabend (1963a, 1965, 1970), and
Hanson (1958) there are suggestions about how differences in conceptualization issue
in differences in perception. For the moment I am solely concerned with the relations
among scientific languages, and with the thesis of conceptual rather than perceptual
incommensurability.
98 The Advancement of Science
appears partial, at best.8 How then do we explain the dialogue between them,
the role that Priestley played in the genesis of Lavoisier's ideas, the assess-
ments and evaluations of Lavoisier's proposals that occur in the writings of
Priestley and his followers? Sensitive to the historical phenomena, Kuhn does
not fully commit himself to the "many worlds" interpretation, offering instead
a position whose ambiguities have excited much subsequent discussion.9
We have a puzzle. If the central terms of Priestley's theoretical language
do not refer, then it seems impossible to ascribe to him the achievements
warranted by the historical record.10 If those terms do refer, then his "world"
must contain entities that we no longer recognize. An adequate solution to
the puzzle must (i) recognize Priestley's contributions to the development of
chemistry, (ii) avoid populating nature with strange entities, (iii) specify the
exact way in which Lavoisier made a conceptual advance.
Begin with a brief overview of features of the language of the phlogiston
theory. This theory attempted to give an account of a number of chemical
reactions and, in particular, it offered an explanation of processes of com-
bustion. Substances which burn are rich in a "principle," phlogiston, which
is imparted to the air in combustion. When we burn wood, for example,
phlogiston is given to the air, leaving ash as a residue. Similarly, when a metal
is heated, phlogiston is emitted, and we obtain the calx of the metal.
Champions of the phlogiston theory knew that, after a while, combustion
in an enclosed space will cease. They explained this phenomenon by supposing
that air has a limited capacity for absorbing phlogiston. By heating the red
calx of mercury on its own, Priestley found that he could obtain the metal
mercury, and a new kind of "air," which he called dephlogisticated air. (Ac-
cording to the phlogiston theory, the calx of mercury has been turned into
the metal mercury by taking up phlogiston; since the phlogiston must have
been taken from the air, the resultant air is dephlogisticated.) Dephlogisti-
cated air supports combustion (and respiration) better than ordinary airbut
this is only to be expected, since the removal of phlogiston from the air leaves
the air with a greater capacity for absorbing phlogiston.
In the last decades of the eighteenth century, the phlogiston theorists
became interested in the properties of a gas, which they obtained by pouring
a strong acid (concentrated sulfuric acid, for example) over a metal or by
passing steam over heated iron. They called the gas inflammable air.
8. See (Kuhn 1962/1970 149, Kuhn 1970 200, 201, 204). Perhaps the most radical
statement of the "many worlds" position is at (Kuhn 1962/1970 102-103).
9. For Kuhn's own retrospective on his views, see his (1983). Discussions and critiques
of his claims about conceptual incommensurability can be found in (Shapere 1964, 1965),
(Scheffler 1967), (Kordig 1971), (Davidson 1973), (Field 1973), (Kitcher 1978), and (Devitt
1981). The present discussion closely follows the line of my earlier attempts in my (1978,
1982, 1983c).
10. Even if one takes Priestley's principal accomplishments to reside in his actions rather
than in his words, there is still a problem in coordinating his dicta with his behavior. If we
assume that his central terms do not refer, then he has to appear as serendipitous
producing a stream of babble while simultaneously doing things (isolating oxygen, syn-
thesizing water) that somehow enable others to repeat his doings and discuss them.
Varieties of Progress 99
11. In the wake of the work of Kuhn and Feyerabend, such simple references to truth
are likely to provoke mirth, scorn, or shock. It is well to remember that part of the reason
for viewing these suggestions as simplistic depends on accepting Kuhnian or Feyerabendian
theses about what occurs in episodes like the transition between Priestley and Lavoisier,
already adopting one of their solutions to our puzzle. Thus a partial defense of our natural
way of talking about past theorists as articulating truths is to show that we can accommodate
the historical material while continuing to engage in that talk. Other parts of the defense
will be supplied in Chapter 5.
12. This consequence strikes me as intuitive, but it does merit some discussion. We
could defend it by supposing (following Frege) that an expression formed by extensionally
embedding a nonreferring expression fails to refer or by adopting a principle of substitut-
ability for nonreferring names: a complex expression formed by extensionally embedding
a nonreferring expression preserves its reference under substitution of nonreferring expres-
sions that belong to the same syntactic category. The general idea of substitutability prin-
ciples for nonreferring expressions is due to Hartry Field.
100 The Advancement of Science
Even a brief reading of the writings of the phlogistonians reinforces the
idea of true doctrines trying to escape from flawed language. Consider Priest-
ley's account of his first experience of breathing oxygen.
My reader will not wonder, that, after having ascertained the superior good-
ness of dephlogisticated air by mice living in it, and the other tests above
mentioned, I should have had the curiosity to taste it myself. I have gratified
that curiosity, by breathing it, . . . The feeling of it to my lungs was not sensibly
different from that of common air; but I fancied that my breast felt peculiarly
light and easy for some time afterwards. (Priestley 1775/1970 II 161-162).
Surely Priestley's token of 'dephlogisticated air' refers to the substance which
he and the mice breathednamely, oxygen.
Similarly, it seems that Cavendish's uses of the terms refer in the following
passage:
From the foregoing experiments it appears, that when a mixture of inflam-
mable and dephlogisticated air is exploded in such proportion that the burnt
air is not much phlogisticated, the condensed liquor contains a little acid
which is always of the nitrous kind, whatever substance the dephlogisticated
air is procured from; but if the proportion be such that the burnt air is almost
entirely phlogisticated, the condensed liquor is not at all acid, but seems pure
water, without any addition whatever. . . . (Cavendish 1783/1961 19)
We readily understand what prompted this description. Cavendish had per-
formed a series of experiments in which samples of hydrogen and oxygen
were "exploded" together; in some cases, the oxygen obtained was not en-
tirely pure, and the small amount of nitrogen, mixed in it, participated in a
reaction to form a small amount of nitric acid; when the sample of oxygen
was pure there was no formation of nitric acid, and the only reaction was the
combination of hydrogen and oxygen to form water. In reporting the exper-
iments he had done, Cavendish used the terms of the phlogiston theory to
refer to the substances on which he had performed the experiments: that is,
he used 'inflammable air' to refer to hydrogen, 'dephlogisticated air' to refer
to oxygen, and so forth.13
13. These claims and my kindred remarks about Priestley will surely jar some historical
sensibilities. I am using the ideas of modern science to provide a picture of the relationships
between Priestley and Cavendish and the world. Isn't this Whiggish, illegitimate, question
begging? I reply that without some picture of the relationship between past figures and the
world, history is impossible: we cannot frame any hypotheses about what their discourse
means. Thus the strategy of agnosticism leaves the historian in the presence of a parade
of uninterpretable symbols, a detached text without significance. So where should we turn
for a view of nature that will enable us to understand the actions and reactions of the
protagonists? Is there some other account that would serve us better than modern science?
Why not the bestthe account we take for granted in planning our own activities?
One modish answer is to insist that we adopt the "actor's categories." To this it might
be pertinent to inquire why we should favor a discredited account of nature rather than
the one which we typically adopt. But there is a deeper point. Without an account of the
relationship between the actor and nature already in hand we cannot interpret the language
of past science, cannot even formulate the actor's categories. I claim that the account I
Varieties of Progress 101
provide in this section, far from riding roughshod over the actor's categories, actually
displays those categories and enables us to see what Priestley and Cavendish meant.
Obviously, the historical sensitivities are raised by genuine worries. We should not
assume that homonymous expressions have the same reference (or reference potential).
We should not assume that the account of nature that we use in interpreting the languages
of the past is infallibleit is not, and the history that we develop on the basis of the picture
offered by present science inherits the troubles of modern science (and possibly more
besides). I hope that thoughtful historians will appreciate the ways in which my use of
present science responds to the points of real concern, and will not be misled into over-
reaction, based ultimately on zeal for an impossible projectthat of some kind of empathic
telekinesis into the minds of past scientists.
14. I should emphasize that Grandy's principle diverges from the principle of charity
in not requiring that we maximize agreement with those whom we interpret. The underlying
idea is that we attribute beliefs by supposing that our interlocutors have cognitive equipment
that is similar to our own, and using what we know about the experiences they have had.
Those who have received very different stimuli can be expected to have formed very
different beliefs; (Grandy 1973) provides a lucid elaboration of these points. I should note
that for some interpretative projectsbut probably not for those which arise in doing the
history of sciencethe principle of humanity will need refining in much the way that it
corrected the principle of charity: on occasion, background evidence may suggest to us that
we are not dealing with interlocutors who share cognitive equipment similar to our own.
102 The Advancement of Science
viz. that, being capable of taking more phlogiston from nitrous air, it therefore
originally contains less of this principle [i.e., phlogiston]; my next inquiry
was, by what means it comes to be so pure, or philosophically speaking, to
be so much dephlogisticated; . . . (Priestley 1775/1970 II120)
From our perspective Priestley has misdecribed the new gas. His remarks
on this occasion identify the gas obtained by heating the red calx of mercury
as dephlogisticated air, and set up a mode of reference for tokens of 'dephlo-
gisticated air' that fixes the reference by the description "the substance obtained
when the substance emitted in combustion is removed from the air." When
tokens are produced, with this mode of reference, they fail to refer. On other
occasions, Priestley and his friends (Cavendish, for example) produce tokens
of 'dephlogisticated air' with a different mode of reference. Their dominant
intention is to refer to the kind of stuff that was isolated in the experiments
they are reportingto wit, oxygen. Thus, in the passages quoted earlier,
Priestley describes the sensation of breathing oxygen and Cavendish proposes
an account of the composition of water. Because the referents of the tokens
of 'dephlogisticated air' are fixed differently on these occasions, Priestley and
Cavendish enunciate important new truths.15
Of course, they do so in a misguided language. But it is not hard to pinpoint
the troubles of their idiom. Not only is the reference potential of terms such
as 'dephlogisticated air' (and even of 'phlogiston'; see Kitcher 1978 534)
heterogeneous, but the modes of reference are connected by a faulty theo-
retical hypothesis. Thinking of combustion as a process of emission, Priestley
and his friends are led to take it for granted that the gas liberated from the
red calx of mercury has been obtained by removing phlogiston from the air.
Their connection of what we see as two distinct modes of reference, that fix
reference to oxygen and to nothing, respectively, rests on their acceptance
of the following hypothesis:
Phlogistonians all accept H, and, in consequence, they are all willing to use
tokens of 'dephlogisticated air' whose references are fixed either by the de-
scription "the substance obtained by removing from the air the substance
emitted in combustion" or by encounters with samples of the gas obtained
by heating the red calx of mercury.
We can now give a precise sense to the claim that a term employed by a
15. Note that applying a pure causal theory of reference to Priestley and Cavendish
misdescribes those occasions (the initial misidentification of oxygen, for example) on which
they fail to refer. The pure descriptive theory misdescribes those occasions (when they are
reporting on samples of the gas that they have prepared) on which they succeed in referring
to oxygen. So, I claim, the appeal to heterogeneous reference potentials is needed to account
for the mixed record of success and failure.
Varieties of Progress 103
There are many situations in which these maxims conflict. When they do,
the scientist "chooses" among themin the sense that there is a dominant
intention to obey one rather than the others. If the choices are made in
different ways on different occasions of utterance, different tokens of the
same type can refer differently.16
Conceptual progress should be assessed in terms of proximity to the ideal
state. One of the goals of science is the construction of a language in which
the expressions refer to the genuine kinds and in which descriptive specifi-
cations of the referents of tokens can be given. For such a language the three
maxims are in harmony.
We can now introduce a notion of conceptual progressiveness:
16. This should be easily understandable in terms of the psychological picture developed
in Section 2 of the last chapter. We can envisage scientists as subscribing to three different
long-term goals. In situations of conflict, one of these goals is activated, and, in conjunction
with other features of working memory it dictates the mode of reference of the term token
produced.
Varieties of Progress 105
(b) for any expression e in C1, if there is a kind to which some token
of e refers, then there is an expression e* in C2 which has tokens referring
to that kind
(c) for any e, e*, [as in (b)], the reference potential of e* refines the
reference potential of e, either by adding a description that picks out the
pertinent kind of by abandoning a mode of reference determination
belonging to the reference potential of e that failed to pick out the
pertinent kind.
In this definition, clauses (a) and (b) are necessary to eliminate the possibility
that refinements of reference potential might be achieved at the cost of ex-
pressive loss. We isolate those parts of the language that do not change [in
(a)], and demand that any kinds that can be discussed in the old language
should be specifiable in the new [in (b)]. Improvements come about by aban-
doning modes of reference that are not in accord with one of the maxims or
by adding modes of reference that would be in accord with both clarity and
naturalism.
Numerous examples from the history of science reveal conceptual progress
in the sense captured in (CP). The shift from Priestley to Lavoisier shows
retention of the ability to refer to oxygen, with replacement of the flawed
reference potential of the term Priestley employed to refer to oxygen ('de-
phlogisticated air') with the refined reference potential of Lavoisier's expres-
sion 'oxygen.' Similarly, in the Copernican revolution, a term, 'planet,' some
of whose tokens previously referred to a natural kind (the set of things we
would count as planets of the solar system) and some of whose tokens referred
to a set that is not a kind (the set whose members are Mercury, Venus, Mars,
Jupiter, and Saturn), underwent refinement of its reference potential. After
Copernicus, reference can be fixed descriptively to a kind (the planets of the
solar system) and the flawed mode of reference to the partial subset is
dropped. Later in the history of astronomy, 'planet' comes to have a reference
potential that shows further refinement: a descriptive mode of reference is
used to fix reference to the kind that includes the planets of all stars, and the
subkind of planets of the solar system is picked out by explicit restriction.
Similar accounts can be provided for the other examples on my original list.
5. Explanatory Progress
Explanatory progress consists in improving our view of the dependencies of
phenomena. Scientists typically recognize some phenomena as prior, others
as dependent. For example, ever since Dalton, chemists have regarded mo-
lecular arrangements and rearrangements as prior to the macroscopic phe-
nomena of chemical reactions, and, since the 1960s, geologists have viewed
interactions among plates as prior to facts about mountain building and earth-
quakes. General ideas about these dependencies can be shared while specific
theses about the details of dependence are debated. Commitment to Dalton-
106 The Advancement of Science
ian atomism in the nineteenth century persists through any number of claims
about the formulas of chemical compounds.
Judgments about dependency can be understood as concerned with the
forms of ideal explanatory texts that the scientists in question envisage. One
component of practice is a collection of patterns or explanatory schemata.
Fields of science make explanatory progress when later practices introduce
explanatory schemata that are better than those adopted by earlier practices.
What does 'better' mean here? One answer, the response of robust realism,
is to declare that there is an objective order of dependency in nature.17 The
character of chemical reactions is objectively dependent on underlying mo-
lecular properties, facts about mountain building and earthquake zones are
objectively dependent on the facts of plate tectonics, the characteristics of
contemporary organisms are objectively dependent on the evolutionary his-
tories of those organisms, and so forth. Recognizing these dependencies
and how to extend them furtheris an important progressive step.
The robust realism outlined here is not the only way to say what is meant
by improving the set of explanatory schemata. An alternative is to tie advances
to enhanced ability to understand the phenomena, for example by proposing
that a particular collection of schemata offers a more unified vision of the
world. I shall postpone comparison of these two metaphysical options until
the next chapter (see Section 8, where I also take up the question of how
to explicate the discussion of kinds that I took for granted in the previous
section). For now it will suffice to see the general shape of the view. Ex-
planatory progress consists in improving our account of the structure of nature,
an account embodied in the schemata of our practices. Improvement consists
either in matching our schemata to the mind-independent ordering of phe-
nomena (the robust realist version) or in producing schemata that are better
able to meet some criterion of organization (for example, greater unification).
Let us consider some examples from the history of science. Almost every-
thing that Dalton maintained about atoms was wrong. Nonetheless, we rec-
ognize his formulation of atomic chemistry as an important progressive step
because it introduced a new correct explanatory schema. Dalton proposed to
explain facts about the course of chemical reactions (and about the weights
of reactants and products) by appealing to premises about atoms, premises
that specify the "fixed proportions" in which atoms of different elements
combine. Chemists ever since have endorsed the claim that this is a correct
picture of the objective dependencies, and they have further articulated Dai-
ton's schema.
17. This conception is as old as the Aristotelian idea of an order of being. For robust
realists, the aim of science is taken to be that of delineating the fundamental mechanisms
at work in nature. If the achieving of adequate concepts is understood in terms of an
anatomical description, the display of objective dependencies corresponds to physiology
showing how things work. For a modern attempt to give literal significance to these meta-
phors, see Wesley Salmon's defense of the "ontic conception" of explanation in his (1984).
I discuss some of the epistemological problems associated with Salmon's approachas well
as some of its considerable meritsin my (1989). See also Section 8 of the next chapter.
Varieties of Progress 107
DALTON
Question: Why does one of the compounds between X and Y always
contain X and Y in the weight ratio m : n?
Answer:
(1) There is a compound Z between X and Y that has the atomic formula
XpYq.
MOLECULAR AFFINITIES
Question: Why does one of the compounds between X and Y always
contain X and Y in the weight ratio m : n?
Answer:
(1) Molecules of X contain j atoms, molecules of Y contain k atoms.
(2) The valences ("affinities") of X and Y are
(3) The possible equations governing chemical reactions between X and
Y are. . . .
(4) There is a compound Z between X and Y that has the atomic formula
XPYq.
(5) The atomic weight of X is x; the atomic weight of Y is y.
(6) The weight ratio of X to Y in Z is px : qy (= m : n).
SHELLFILLING
Question: Why does one of the compounds between X and Y always
contain X and Y in the weight ratio m : n?
Answer:
(1) Molecules of X contain j atoms; molecules of Y contain k atoms.
(2) (a) An uncharged atom of X contains g electrons; an uncharged
atom of Y contains h electrons.
(b) The first incompletely filled shell level around an uncharged
atom of X contains g1 electrons and g2 free places, the first incompletely
filled shell-level around an uncharged atom of Y contains h1 electrons
and h2 free places.
(c) The combinations of atoms of X and Y that will achieve shell
filling either through electron transfer (ionic bonding) or through electron
sharing (covalent bonding) are . . . .
(d) The valences ("affinities") of X and Y are
(3)-(6) As for MOLECULAR AFFINITIES.
18. In the example from the history of chemistry I have been concerned to draw the
line of development without introducing the historical details. The Darwinian example of
Chapter 2 is meant to balance this by providing much more of the scientific background
and linking the schemata to specific points in the history of biology. With the discussion
of this section in mind, it should be possible for the reader to return to the more discursive
treatment of Chapter 2, identifying the same structural relations and progressive picture
that I display here. Another example, treated in more detail in (Kitcher 1989), focuses on
the development of explanations of hereditary distributions from Mendel to Watson and
Crick.
110 The Advancement of Science
of chemical reactions (specifically about the weights of reactants and products)
on the facts about atomic combination and by Darwin's insight that distri-
butions and relationships among contemporary organisms are dependent on
the course of descent with modification. Second, we have the elimination of
incorrect schemata, such as Darwin's appeals to the inheritance of acquired
characteristics. Third, we find the generalization of schemata, rendering them
able to deal correctly with a broader class of instances: evolutionary theorists
who appeal to classical individual selection alone are correct in identifying a
certain type of dependence, but their proposals are less generaland there-
fore less completethan those which allow for drift, migration, meiotic drive,
inclusive fitness effects, developmental constraints, and so forth. Finally, there
is explanatory extension, when the picture of dependencies is embedded
within some larger scheme. The incorporation of Darwin's selectionist pat-
terns within NEO-DARWINIAN SELECTION and the embedding of atomic chemistry
in quantum physics show this process at work.
At the end of Chapter 2 I suggested that the history of evolutionary biology
does not show an indecisive alternation but something like cumulative prog-
ress. The goal of this section and its predecessor has been to make that
judgment more precise. There do seem to be cumulative processes of spe-
cifiable kinds in my examples, as schemata are introduced, refined, gener-
alized, and extended.
But how typical are these examples? Am I cheating by simply drawing on
periods that Kuhn would classify as belonging to "normal science," and that
other students of scientific change would rank in analogous categories?19 My
reply consists of a historical reminder and a promise. The historical reminder
is to note that the periods over which I have traced the development of
schemata not only are long (of the order of 100 to 200 years at the stage of
most rapid change in the histories of the relevant sciences) but also span
"normal" and "revolutionary" episodes. In each case, even if all schemata
after the first can be assigned to a single "normal scientific" tradition, the
introduction of the first marks a progressive step over the accounts previously
available: the schemata introduced by Dalton, Mendel, and Darwin provide
answers to the questions they address that identify correct dependencies un-
appreciated by their predecessors. Moreover, if one wants to extend the notion
of "normal science" (or some equivalent) to cover the entire periods through
which I trace the refinement, generalization, and extension of schemata, then
it will be necessary to downplay the significance of "revolutions" in science.
These periods are so large, and the processes I identify so prevalent, that
"normal science" will be everywhere once a field attains maturity.
But I do not want to claim too much for these examples. They serve to
remind us of what scientists are sometimes brave enough to say, that domi-
nation of philosophical discussion by the vicissitudes of a few concepts from
19. These questions were posed forcefully to me by Larry Laudan and Rachel Laudan.
I am grateful to both of them for much helpful discussion of this point, although I suspect
that neither will be convinced by my response.
Varieties of Progress 111
20. Thus Mayr protests the influence of a few theories from physics on the development
of an allegedly general picture of science and scientific progress (1982 857). In a valuable
review of Mayr's book, John Maynard Smith is admirably forthright: "Unfashionable as it
may be to say so, we really do have a better grasp of biology today than any generation
before us, and if further progress is to be made it will have to start from where we now
stand" (see Maynard Smith 1989 11). I hope that my account explains clearly why Mayr
and Maynard Smith are right.
112 The Advancement of Science
6. Derivative Notions
Significant questions arise against the background of the ordering of the
phenomena captured in our explanatory schemata. There are two main ways
in which significance can accrue to a question, generating application questions
and presuppositional questions, respectively. Application questions are gen-
erated from projects of finding particular instantiations of the available sche-
mata. Presuppositional questions investigate the conditions that must obtain
if available schemata are to be instantiated.
In the early stages after a new schema has been introduced into consensus
practice, almost all instantiations of it are important. Thus, in the early nine-
teenth century, chemists committed to Dalton's atomic theory regarded any
problem of understanding the weight relations among compounds as signifi-
cant. Because solutions to these problems required them to make judgments
about the chemical formulae of compounds, the question of how to assign
chemical formulae and to discriminate among proposed formulae obtained
derivative significance. Similarly, in the years after the publication of the
Origin, naturalists set to work to find convincing instantiations of Darwin's
21. My view of explanatory progress plainly has some affinity with the ideas of the
logical empiricists about the unity of science. However, there are important differences
stemming from my rejection of the demand that there be accumulation at the level of details.
For more exact discussion of the relationship, see (Kitcher 1984, 1989), which explore the
question in the context of the growth of genetics.
The writings of Lakatos, Laudan, and Toulmin, all of whom hope to specify broader
units (research programs, research traditions) behind individual theories are sensitive to
the problems that beset the demand for accumulation at the level of details. But the emphasis
on empirical prediction and the shying away from analysis of the explanations offered within
a field of science at a time seem to me to interfere with the appreciation of the possibility
of accumulative explanatory structure. Toulmin's conception of "ideals of explanatory
order" comes closest to recognizing what I regard as the salient point: the need to break
away from concentration on accepted statements (a feature of logical empiricism that survives
in Lakatos and Laudan) and to focus on the ways in which statements are used in answering
questions.
Varieties of Progress 113
22. There are filiations here not only to Kuhnian ideas but also to Larry Laudan's
discussion of conceptual problems in his (1977).
114 The Advancement of Science
atoms to satisfy the conditions of the Bohr model and generates the question
of understanding how atoms meeting these conditions can avoid the kinds of
collapse that would be expected from the principles of classical electromag-
netic theory.
In these examples, the presuppositions of the schemata are problematic
in that there are apparently cogent arguments from plausible premises, avail-
able to proponents of the practice, that seem to show that the presuppositions
are false. What are required are a demonstration of the possibility of the
problematic presuppositions and a diagnosis of the flaw in the apparently
persuasive reasoning. Combining this idea with the earlier discussion of ap-
plication questions, we can say that questions are intrinsically significant when
(a) answers to them would exhibit the possibility of instantiating an accepted
schema, or (b) would exhibit the possibility of instantiating an accepted
schema in apparently problematic instances, or (c) would show the possibility
of some problematic presupposition of an accepted schema. Questions are
instrumentally significant when answers to them would answer some intrin-
sically significant question of some other field or answer some instrumentally
significant question of some other field. (The last clause allows for the pos-
sibility of a chain of fields each of which turns for help to its neighbor; but,
at the end of the chain, there must be some field with an intrinsically significant
question.)
The account of significance I have offered is best seen as outlining the
idea of an apparently significant question. Relative to a set of schemata certain
questions should be regarded as significant because they demand instantiations
for those schemata, or instantiations in apparently difficult cases, or dem-
onstrations of the possibility of problematic presuppositions. We can say that
a consensus practice is erotetically well grounded if the questions to which it
assigns significance are indeed those that are significant relative to its sche-
mata. Genuinely significant questions are those that are significant (in the
sense I have indicated) relative to correct schemata. We make erotetic prog-
ress when we have an erotetically well-grounded consensus practice in which
we pose genuinely significant questions that were not previously asked.
Scientists often describe some fieldsespecially those that are regarded
as most excitingby suggesting that they now know how to pose the right
questions. A concept of erotetic progress should capture such descriptions.
Sometimes erotetic progress can be a by-product of conceptual progress:
Priestley's questions about the role of dephlogisticated air in various reactions
are better formulated as questions about oxygen. On other occasions, as I
have suggested in the last paragraph, incorporation of new explanatory sche-
mata generates new genuinely significant questions: a striking example here
is Darwin's initial introduction of his evolutionary schemata, which gave rise
to a host of new application and presuppositional questions. However, there
is a further facet of erotetic progress that has not yet been made explicit. We
make progress by posing more tractable questions.
Scientific fields typically begin with big, vague questions. The introduction
of explanatory schemata structures our large, cloudy wonderings, by sug-
Varieties of Progress 115
23. For an illuminating discussion of the programs that descend from Newton, see
(Schofield 1969). I have sketched an account of these programs in the terms of the present
chapter in (Kitcher 1981 section 3). See also (Boscovich 1763/1966 and Kitcher 1986). In
contemporary developmental biology, there is similar uncertainty about how to focus the
big, vague question, How do organisms develop?
24. For an example of this hierarchy, see (Culp and Kitcher 1989). Obviously, there is
a sense in which the accepted schemata of a practice play a similar role to the "hard core"
of Lakatos or the "core assumptions" of Laudan. However, as I have already noted, one
missing feature of the approaches of Lakatos and Laudan is their concentration on accepted
statements rather than the ways in which accepted statements are used. A consequence of
this seems to me to be that their views of science do not relate to much of the work in
which scientists actually engage. They do not provide accounts that enable us to see how
the very local and specific projects that occupy almost all scientists at almost all times come
to be valued. Here I take Kuhn's analysis of normal science to be suggestive, and I have
tried to articulate the suggestions.
116 The Advancement of Science
appear bothersome. Close attention to my account of correctness of explan-
atory schemata and to the implications of (EP) will reveal that it is possible
to have practices P1, P2 such that neither the transition from Pl to P2 nor the
transition from P2 to P1 would count as explanatorily progressive. For, while
Pl may lack a correct explanatory schema present in P2, it may also contain
a correct explanatory schema absent in P2. Nor is this merely a logical pos-
sibility. Devotees of the alleged phenomenon of "Kuhn loss" will insist that
major shifts in science often involve the abandonment of explanatory insights
that later developments in the field will reestablish. While Newton introduced
the idea of deriving conclusions about accelerations from premises specifying
the forces acting, he gave up the Aristotelian grounding of forces in the
geometrical structure of space, an explanatory proposal that was later re-
covered by Einstein. Lavoisier's new chemistry sacrificed the possibility of
accounting for the combustibility of the metals, which had been successfully
undertaken by proponents of the phlogiston theory and which would be re-
captured in the twentieth century accounts of chemical bonding. I shall not
pursue these examples in detail here (see 9 of Chapter 5 and 9 of Chapter
7), but simply outline a solution to the problem of understanding the shifts
from Aristotle to Newton and from Priestley to Lavoisier as progressive.
Suppose we grant that Aristotle had a correct schema that was abandoned in
Newtonian practice, and that Newton introduced a new correct schema that
Aristotle had lacked (and similarly for Lavoisier and Priestley). From the
perspective of (EP), then, Aristotle and Newton (Priestley and Lavoisier)
cannot be ranked for progressiveness.25 Nonetheless we may still score a
considerable advance in erotetic terms by recognizing that the significant
questions that are abandoned in the shift from Aristotle to Newton are in-
tractable while that transition introduces a host of new significant questions.
Indeed, we might even maintain that the possibility of decomposing the im-
precise (but significant) Aristotelian questions requires the prior posing of
the tractable (and also significant) Newtonian problems.26
The view I have sketched can easily be motivated by using the homely
analogy that I introduced in dealing with explanatory progress. The task of
arriving at a correct account of the structure of nature may require us some-
times to abandon primitive explanatory insights (which will much later be
reinstated in articulated form), just as, in solving certain kinds of puzzles
(particularly those of building three-dimensional models) pieces that actually
fit together, and will eventually be rejoined, may have to be taken apart to
permit the assembly of large configurations. If the sympathetic accounts of
25. Since I take progressiveness to be a relation between practices, and since practices
are multidimensional, it is possible that rival practices could be incomparable on some
dimensions but that one was superior to its rival along others.
26. Thus, if one viewsas Kuhn (1970 206-207) seems tothe Aristotelian approach
as containing an embryonic schema for addressing questions about forces, one that traces
forces to underlying spatial structure, then it is quite pertinent to suggest that development
of this insight only became possible through the detailed mathematical investigations begun
by Galileo and Newton.
Varieties of Progress 117
the achievements of Aristotle and Priestley are correct, then this approach
seems to me to offer a satisfying diagnosis of what was progressive in the
transitions to Newton and Lavoisier, respectively. The losses (if any) were
vague insights that could not be articulated at that stage in the development
of science; the gains, in both instances, were correct explanatory schemata
that generated significant, tractable, questions, and the process of addressing
those questions ultimately led to a recapturing of what was lost. Even if these
transitions do not exhibit explanatory progress, they show erotetic progress.27
Once we have the concept of a significant question at our disposal, it is
relatively simple to understand progressiveness with respect to other com-
ponents of practice. Instruments and experimental techniques are valued
because they enable us to answer significant questions.28 One instrument (or
technique) may do everything another does and more besides. If so, then we
make instrumental (or experimental) progress by adopting a practice in which
the former instrument (technique) replaces the latter. Making this conception
of progress precise requires us to look more carefully at progress in the set
of accepted statements: if we know what counts as improving the set of
accepted statements, then we can characterize instrumental and experimental
progress by recognizing the increased power of instruments and techniques
to deliver improved statements.
We make progress with respect to the set of accepted statements in a
number of ways. Sometimes, scientists eliminate falsehood in favor of truth,
abandon the insignificant, add significant truths, or reconceptualize already
accepted truths. Moreover, as I shall suggest in the next section, they often
replace statements that are further from the truth with those that are closer
to the truth. I shall start with the apparently naive idea that part of scientific
progress consists in accepting statements that are both significant and true.
27. It is worth explicitly forestalling a confusion here. I claimed earlier that explanatory
progressiveness was fundamental and erotetic progressiveness derivative. How then can
one have erotetic progress without explanatory progress? The answer requires a distinction
of levels. To make sense of the notion of erotetic progress, we need the concept of a
significant question. The ultimate bestowers of significance are the correct schemata of a
practice. Thus the general notion of erotetic progress is conceptually dependent on that of
explanatory progress (more precisely on notions that figure in the analysis of explanatory
progress). However, it is quite possible that, in specific cases, loss of a correct schema that
cannot yet be instantiated should go hand in hand with the acquisition of the ability to
pose, for the first time, questions that are significant and whose significance accrues from
correct schemata (either those that are retained or some that are introduced in the new
practice). Moreover, these questions may even pave the way for future refinement and
instantiation of the now-discarded correct schemata. In either case, the transition will show
erotetic progress without explanatory progress.
28. I owe to Gareth Matthews the observation that, while some of my relations of
progressiveness concern the attaining of certain epistemic desiderata, others involve achiev-
ing the means for going further. Thus, if we are making conceptual or explanatory progress
we have already gathered some good things (like the firm that is already making a profit).
If we are making instrumental or erotetic progress then we have prepared ourselves well
for gathering good things in the future (like the firm that has invested wisely in new
ventures).
118 The Advancement of Science
Philosophical reticence about the attainment of significant truth is the
result of a failure of nerve, induced by thinking about the problem in a faulty
way. Beguiled by the notion that scientific significance accrues only to systems
of generalizationstheories, conceived as axiomatic deductive systemsthe
problem is framed first in terms of the truth of very general axioms. A pes-
simistic induction on the history of science (the product of reflection on the
famous grand generalizations of our predecessorsnotably classical physi-
cists) instills a conviction that these axioms cannot be strictly true.29 So the
best we can hope to achieve is, apparently, scientific systems that are "close
to the truth." Popperian moves to find measures of truth content become
inviting at this juncture, and we are plunged into a technical morass of com-
paring the "sizes" of infinite sets of true and false consequences.
Tradition takes a misguided view of significance and so brings truth into
the picture in far too ambitious a way, framing the crucial issues in terms of
truth for whole theories. My approach circumvents these difficulties by offering
a quite different view of scientific significance. A significant statement is a
potential answer to a significant question. What we strive for, when we can
get them, are true significant statements, that is, true answers to significant
questions. Sometimes, and with differing frequencies in different sciences,
those statements are universal generalizations. However, there are important
fields of science in which exceptionless generalizations are not sought.
Consider some examples. "DNA molecules consist of two helical strands,
wound around one another in opposite directions"is this statement true?
is it a universal generalization? what makes it significant? Molecular biologists
(and many other scientists) would surely count the statement as among the
most significant truths enunciated in the past half century (to be conservative
about the time period). But if the statement is construed as a strictly universal
generalization it is immediately clear that it is false. Not all DNAs are found
in double helical form: contemporary molecular biology makes considerable
use of single-stranded DNA. The existence of such molecules is not at all at
odds with the intended interpretation of the original statement, which offered
a restricted generalization about the DNA molecules typically found in cells
in nature throughout most of the phases of the life of the cell. The restricting
conditions are not formulated precisely, and, I suspect, nobody knows how
to formulate them precisely. Nevertheless, the statement is both significant
and true: significant because it answers the significant question, What is the
structure of the genetic material?, true because counterinstances such as the
one I have mentioned fall outside its intended scope.
Sometimes significant statements are even more obviously particular. Con-
sider (1) "Part of Western California is at the junction of two plates that slide
past one another," (2) "Gangs of male Florida scrubjays are sometimes able
to expand the territories in which they reside." Such statements obtain their
significance from the role they play in answering significant questions. (1) is
an important part of an answer to the question, Why is there an earthquake
zone in western California? (a question of cognitive significance for some, of
practical significance for others, and of both kinds of significance for some
of us!). Similarly, (2) plays a role in answering the question, Why do male
Florida scrub jays often return to the parental nest and assist in the feeding
of siblings?since, as detailed research on these birds makes clear, there are
beneficial consequences for inclusive fitness not only in providing relief for
overworked parents but also in increasing the possibility of gaining a territory
by imperialist expansion (see Woolfenden and Fitzpatrick 1984).30
Is it reasonable to believe that all the particular statements that fill con-
temporary scientific journals will endure as parts of the consensus practices
of future fields of science? No. But the main problem is not with truth but
with significance. If you turn to the pages of Nature (or other scientific jour-
nals) fifty years ago, they will present conclusions that are sometimes, by our
lights, oddly formulated but substantially correctconclusions, however, that
no longer seem to be significant. Many of them never became part of consensus
practice (or at least part of the consensus practice of a significant subdisci-
plinary group). Others enjoyed a brief career in consensus before being
discarded.
Why does this occur? Because the explanatory schemata of a practice
generate significant primary questions, which spawn derivative questions: an-
swering an appropriate sequence of the latter is seen as a way to address the
former. Not all routes succeed. A community striving to answer Q formulates
the strategy of doing so by tackling q1, q2, and q3 (in that order). The initial
answer to ql is hailed as significant, but then the endeavor becomes stuck.
Ultimately another route to answering Q is found and successfully exploited,
and the answer to q1 disappears from consensus practice. Moreover, even
attaining the answer to a primary significant question by solving some deriv-
ative problems may not endure as a permanent achievement. The same work
may be done more decisively or more elegantly by later scientists. Consensus
practice is economical. The exemplary studies of one generation can be dis-
placed by better exemplars.31
30. Examples can be multiplied by looking at any issue of any scientific journal.
31. Thus, at a particular point in time, the hierarchy of significant question forms
corresponds to numerous hierarchies of significant particular questions. Scientists often
work on specific projects in the hope that the instances they address will prove the key to
tackling the question forms. Although they may reap all kinds of successes, their own
favored instances may vanish from the subsequent discussions of the field simply because
some rival hierarchy of specific projects provides a clearer or more elegant treatment of
the question form.
However, as Rob Cummins pointed out to me, it is possible that work done on one
sequence of questions should fail to be adopted as the definitive treatment of the ultimate
issue but should prove useful in some other area of research. So, for example, some research
efforts channeled originally toward problems within physics are attributed enduring sig-
120 The Advancement of Science
7. Verisimilitude
The friends of verisimilitude rise up to protest: not all scientific state-
ments are true. They are right. Moreover, we sometimes make progress by im-
proving the statements we accept, even though we do not attain truth. For
all my efforts to avoid it, are we not finally stuck with the old problem of
verisimilitude?
That problem has two parts. The first part concerns the search for
exceptionless generalizations. Some sciences do offer exceptionless gener-
alizations, and truth for such statements is typically harder to come by.
However, there are some circumstances under which we naturally rank later
generalizations as closer to the truth than earlier ones. Consider, for ex-
ample, the bare neo-Darwinian statementpopular in the early days of
critical thinking about group selectionthat alleles disposing their bearers
to forms of behavior that would aid others at costs to themselves would
be selected against. For present purposes we can focus on two complicating
conditions: (i) if the allele is associated with a disposition to favor only (or
primarily) kin, it may be maintained in the population through inclusive
fitness effects; (ii) if there is a structured population in which groups found
descendant groups at different rates, dependent on internal conditions that
are fostered by the presence of altruists, then, provided that the group
founding process goes forward quickly enough, the altruistic alleles can be
maintained.32 Now we can formulate three generalizations: "Altruistic alleles
are always opposed by natural selection," "Altruistic alleles are always
opposed by natural selection except when there is an offsetting inclusive
fitness effect," "Altruistic alleles are always opposed by natural selection
except when there is a sufficiently strong effect from population structure
nificance because of their import for resolving debates within biology and geology about
the age of the Earth. Significance can sometimes be serendipitous.
32. The first type of exceptional case is, of course, treated in Hamilton's classic pair of
papers on inclusive fitness (1964). The second emerges from Maynard Smith's criticisms of
group selection (1964) and has been further developed by a number of writers (see Rough-
garden 1979 for a survey, and D. S. Wilson 1980, 1983 for the most general treatment;
Sober 1984 provides relatively nontechnical expositions of the main ideas).
Varieties of Progress 121
33. The reason is quite straightforward. The conditions for Miller's argument are not
met in this case because the later generalizations are restrictions of earlier ones that avoid
some of the original false consequences.
34. The process of "mopping up the exceptions" in science is typically not just a matter
of restricting the scope of the generalization. This is clear in my original example, in which
one does not simply declare that cases in which there are inclusive fitness or group structure
effects are off limits but presents an analysis of how these factors would affect the process
of selection. Scope restriction is only tolerable when the abandoned instances are of no
significance.
122 The Advancement of Science
can ask whether the set of accepted statements of one practice (Fresnel's) is
progressive with respect to the set of accepted statements of another practice
(Brewster's): that inquiry might be undertaken by considering whether Fres-
nel replaced significant falsehoods or insignificant truths with significant
truths; or, using the ideas presented in this section, we might consider whether
Fresnel's magnitude ascriptions were closer to the actual values than Brew-
ster's. The account of progress I am offering is thus keyed to investigating
the, fine structure of what Fresnel and Brewster wrote and said (and what they
didfor we can compare the instruments and experimental techniques they
employed) and to resisting the comparison of slogans. "Light is a wave" and
"Light is a particle" are, at best, gross advertisements for the two practices,
slogans to be traded in the opening moments of debate.
"Difficult cases" can easily be manufactured for an account of progress
(such as mine) by pressing the concept of verisimilitude where it should not
be applied, or, more generally, insisting that progressiveness be gauged along
some particular dimension. My account of verimisilitude does not allow us
to rank "Light is a wave" versus "Light is a particle." Nor should it, for there
is no plausibility in the claim that one of these is closer to the truth than the
other. But my total account of progress does allow us, to assess the practices
of Fresnel and Brewster, enabling us to see that Fresnel made conceptual and
explanatory advances, that he was able to answer correctly significant ques-
tions that Brewster could not (questions about interference and diffraction,
for example).37 We can loosely sum up all these advances by suggesting that
the wave theory propounded by Fresnel was closer to the truth than Brewster's
corpuscular theorybut this is only shorthand for a complex of relations of
the types I have been at pains to characterize throughout this chapter, not a
claim about the restricted notion of verisimilitude that figures in part of the
story.
In the light of this section we can refine the account of instrumental and
experimental progress. Very frequently in the daily practice of a science,
there is a significant question of the form, What is the value of x for S? (What
is the age of that stratum? How many neurons project from this nucleus into
the telencephalon?) There are instruments and techniques designed to answer
classes of such questions. An instrument or technique improves on an earlier
instrument or technique if the values that it generates are closer to the actual
values than those given by the earlier experiment or technique. But is it not
possible that an instrument or technique could yield better results in some
instances but worse results in others? Of course. Geologists know very well
that certain kinds of isotopes are better for dating some types of rocks, other
isotopes more reliable in other cases. The experimental practice of radiometric
dating uses a motley of techniques precisely because there is no all-purpose
decay process that can be used on all rock samples. Geologists combine
separate approaches to fashion a mixture of techniques that will apply across
37. For illuminating discussions of the work of Fresnel and the acceptance of his ideas
about light, see (Worrall 1978, 1989).
124 The Advancement of Science
the entire domain they hope to investigate. The claim that experimental
practice is progressive rests on the idea that the mixture of instruments and
techniques now employed yields in each case a value that is at least as close
to the actual value as that previously generatedor, if there are exceptions,
that these are either rare, insignificant, or both.
8. Two Refinements
The account offered so far is deficient in two main respects. First, as noted
in Section 1, I have sidestepped problems posed by the splitting and merging
of fields. Second, in the last two sections I have overemphasized the goal of
attaining truth (or of improving false statements) and neglected the important
role played by idealizing theories. The present section will attempt to set
these defects right.
When fields of science split, merge, or hybridize, comparisons among
practices cannot properly be made by considering only a single ancestor and
a single descendant. Evidently, if we compare an earlier undifferentiated
practice of chemistry with a later practice of inorganic chemistry, we shall
expect there to be losses. The obvious remedy is to consider total practice at
the earlier and the later times. We can construct a map of the state of science
at a time, and, for each ancestral field, identify its successor fields at the later
time. Then we assess progress within a field by seeing whether the combined
practice of the successor fields is progressive with respect to the practice of
the ancestral field. Alternatively, we can take an ancestor-descendant unit to
be generated in the following way: Start with some ancestral field and identify
all its successors. Now consider all those fields which are ancestral to some
one of the successors already picked out. Next, take each of the ancestral
fields already singled out in the unit, and include all successors. Iterate the
process until no new fields are introduced. Progress is then assessed within
ancestor-descendant units by comparing the combined practices of the suc-
cessor fields in the unit with the combined practice of the ancestral fields.
But this does not yet address one of the most significant aspects of the
splitting and merging of fields, to wit the fact that the large-scale organization
of scientific activity itself reflects a conception of the order of nature and that
changes in this conception may be progressive. Here (as so often) Kuhn is
suggestive (1977 31-65), using the histories of the physical sciences to show
how some changes are not so much modifications within existing fields as
redrawings of the map of science.
We can accommodate the idea of organizational progress, conceived as
improvement of the accepted relations among the sciences, by extending our
view of consensus practice, of conceptual, explanatory, and erotetic progress.
I shall suppose that the consensus practice of a field includes some claims
about the relationship of the phenomena of the field to phenomena studied
by other fields, and that, in the fashion discussed in Section 11 of the last
chapter, we may combine these partial views to a broad consensus vision of
Varieties of Progress 125
the structure of nature. This broad consensus vision underlies the organization
of scientific inquiry, recognizing certain problems and projects as connected
with one another and granting authority to a subgroup of the community
which tackles the problems and pursues the projects. Organizational changes
may redraw boundaries, modify ideas about the dependencies of fields on
others, or incorporate methods of one field into those of another. All these
kinds of change can contribute to varieties of progress that have already been
described. The redrawing of boundaries may constitute conceptual progress,
revision of claims about dependency can be explanatorily progressive, and
integration of concepts and methods from one discipline sometimes provides
strikingly more tractable versions of old problems, thus counting as erotetic
progress. Examples are ready to hand, in Maxwell's unification of the theories
of electricity, magnetism, and light; in the incorporation of the theory of the
chemical bond within atomic physics; and in the development of economic
models in ecology.
Let us now turn to the second deficiency that I diagnosed earlier. One of
the most obvious features of some sciences (notably parts of physics, but also
subdisciplines of ecology and evolutionary biology) is their employment of
idealizations. The generalizations of phenomenological thermodynamics, of
the kinetic theory of gases, of statistical mechanics, are not, strictly speaking,
true.38 Nonetheless, it is entirely legitimate to hold that physics made progress
by first achieving the phenomenological generalizations, then those of kinetic
theory, then those of statistical mechanics. Using the ideas of the last section,
we can recognize the progress of thermodynamics by suggesting that successive
generalizations specify functions that are closer to those actually involved in
the dependencies of various quantities (pressure, volume, and temperature,
for example). There is nothing wrong with thisas far as it goes. However,
it fails to recognize a certain aspect of the progress of thermodynamics: the
goal of the enterprise is not to advance a set of generalizations that are exactly
true.
Start with a distinction. The statements of phenomenological thermody-
namics, of kinetic theory, and of statistical mechanics are rightly counted as
false when we interpret the terms that occur in them as referring to actual
magnitudes of actual gases or of actual molecules. However, if we conceive
of the referents of those terms as fixed through stipulationa molecule of an
ideal gas is to be a Newtonian point particle that engages in perfectly elastic
collisions, and so forththen the theories in question can be seen as simply
elaborating the consequences of these stipulations. To use an old (and dis-
reputable) terminology, the statements are true by convention. As I have
argued elsewhere, when Quinean morals are properly understood, there is
room for truth by convention (see Quine 1966, Kitcher 1983 chapter 4). What
38. For a penetrating exploration of this claim, see (Cartwright 1983). My own discus-
sion of idealization in what follows is brief and plainly needs supplementing by attending
to the important distinctions that Cartwright makes. For present purposes, however, I am
concerned to show how the elaboration of idealizations can be accommodated within my
general approach to cognitive progress.
126 The Advancement of Science
127
128 The Advancement of Science
can be blocked, it remains to show that an account of scientific knowledge
can be developed.
As will become apparent, there are many presentations of critiques of
realism in which different kinds of objections blend. A criticism that begins
by questioning the coherence of realism can shift to a demand for under-
standing how we can know that what scientists say is true. A specific objection
to the possibility of certain kinds of knowledge may conclude with a challenge
to provide a full account of knowledge. This chapter will take up criticisms
of the first two kinds, endeavoring to defend the coherence of the realist
concepts I employ and to turn back particular attacks on the possibility of
attaining truth. The more ambitious task of advancing an account of scientific
reasoning and scientific knowledge will occupy the next three chapters. Be-
cause of the blurring of boundaries among the types of objection, the treat-
ment of some topicsfor example, worries about underdetermination and
theory-ladennesswill be incomplete. I can only promise that more will come
later.
The challenges I shall consider can be distilled into a sequence of questions.
(1) Is it coherent to suppose that the sciences aim at and attain the truth? (2)
How can we know that we are making progress in any of the senses that I
have delineated? (3) Doesn't induction on the past history of the sciences
reveal to us that our current beliefs are overwhelmingly likely to be wrong?
(4) Doesn't the history of the sciences show that the aims that scientists set
for their enterprise vary from period to period? (5) Can I avoid a naive realism
that takes us to have direct access to the facts about nature, unmediated by
background theory? (6) Isn't my account subverted by the fact that our system
of beliefs (including beliefs about our access to nature) results from inter-
actions among people who "negotiate" what is to be accepted? (7) Does that
account lead to an untenable form of realism in which it is assumed that
nature has a determinate categorial and causal structure, the kind of mind-
independent structure that has been dubious ever since Hume? I shall take
up these objections in turn.
2. Rehabilitating Truth
Truth can seem commonplace, or utterly mysterious, depending on perspec-
tive. One type of concern about the kind of account of scientific progress I
outlined in Chapter 4 proceeds by making appeals to truth look grandiose,
even mystical. It is well to start by reminding ourselves of humbler uses of
the idea.
Semantic facts concern the relation between language users and nature.
In virtue of the state of the language user and the state of the rest of the
world, there is sometimes a relationthe relation of referencebetween the
words spoken or written and items in the world. In consequence, the statement
represents the world as being some particular way. The statement is true just
in case the way in which the world is represented is the way it really is.
Realism and Scientific Progress 129
3. This project is a preliminary to answering questions that have been forcefully posed
by Kuhn and many of his predecessors and contemporaries: how do we justify the claims
that certain representations match what is "really there"? Attention to the worries raised
by Popper, Duhem, Quine, and others will come later. But, as Popper himself clearly saw,
the first issue to settle is the sense of realist appeals to truth by correspondence. My approach
to this question will follow Popper's own lead, building on the work of Alfred Tarski (1936,
1944). (See Popper 1959 274 fn. *1; my own discussion is considerably indebted to Field
1972.)
130 The Advancement of Science
and predict the behavior of our fellows by attributing to them states with
propositional content (for celebration of this theme, see Dennett 1987). We
explain and predict the differential successes of our fellows in coping with
the world by supposing that there are relations between the elements of their
representations and independent objects. Those with correct beliefs about
spatial relations can navigate their way more successfully than those who have
faulty beliefs, and they can do so because their beliefs correspond to the ways
in which the constituents of the local environment are arranged. Simple psy-
chological ideas form part of the explanation of everyday behavior. Simple
semantical ideas, added to those simple psychological ideas, explain the con-
sequences of behavior.6
Correspondence truth, I suggest, begins at home. Few are born antireal-
ists, and those who achieve antirealism typically do so because it is thrust
upon them by arguments they feel unable to answer. In the present instance,
the arguments derive from the denial of the theory-independent perspective.
Kuhn's sympathy for this line of criticism is suggested in his phraseology:
"There is no theory-independent way to reconstruct phrases like 'really there';
the notion of a match between the ontology of a theory and its 'real' coun-
terpart in nature now seems to me illusive in principle" (1962/70 206). Now
it is, of course, right to insist that any description of what is "really there,"
however abstract we try to make our formulation, will presuppose some
language, some conceptualization of nature. Equally, if one thinks of a match
between cognitive/linguistic items and independent nature as requiring some
possible process of matching, then, because there is no Archimedean point
from which both sides can be viewed, the notion of match will come to seem
"illusive." The common root of both ideas is that we have no access to nature
which does not involve some elements of some scientific practicea point
which is uncontroversial once one has recognized that the categories and
beliefs of common sense are themselves parts of primitive scientific practices.7
But why should this doom the idea that there is something independent of
us to which we have access through processes that are dependent on the state
of current science/common sense?
could be recast in the terms that the critic favors for talking about truth and reality (the
"internal realist" idiom in Putnam's case). But, I believe, such attacks can be turned back
by taking seriously the idea that semantic relations can be identified with thoroughly natural
(physicalist) relationsthat is, working out the program advertised in (Field 1972). For
the present, that must simply remain a promissory note.
6. It is not hard to see that this perspective on the role of semantics in everyday practice
foreshadows the popular argument for scientific realism in terms of the successes begotten
of our scientific practices. For more discussion of this see 3-5 of this chapter.
7. This theme is even more apparent in the writings of Feyerabend than of Kuhn (see
Feyerabend 1963a, 1963b). The emphasis on common sense as itself a primitive theory is
sharply directed against the partisans of "ordinary-language" philosophy (Feyerabend 1963a
fn. 99 expresses this clearly and wittily). For many philosophers of science, the point hardly
came as a shock. Popper, for example, had already detected theoretical involvement in
our most basic, commonsensical observational claims (1959 111).
132 The Advancement of Science
The move from the theory dependence of our perception of nature to the
theory-dependence of nature is apparent in Kuhn's discussion of Priestley and
Lavoisier.
At the very least, as a result of discovering oxygen, Lavoisier saw nature
differently. And in the absence of some recourse to that hypothetical fixed
nature that he "saw differently," the principle of economy will urge us to
say that after discovering oxygen Lavoisier worked in a different world.
(Kuhn 1962/70 118)
Realists have their preferred picture of what happened to Lavoisier. There
was a constant fixed nature. As a result of a series of interactions with this
nature, Lavoisier's cognitive system underwent some changes (the changes
we describe by saying that he acquired the concept of oxygen, came to believe
that combustion involved the absorption of oxygen, and so forth). In con-
sequence, his subsequent interactions with nature activated new propensities
and induced PI beliefs which would not have been induced in his former self
(or in his friend Priestley). Realists add that Lavoisier's new beliefs are better
than his old ones because, in some instances, they match the way nature is.
Kuhn's challenge to this story seems to be that its postulation of a fixed theory-
independent nature is otiose. Since we have no independent access to nature,
we are just making up an unnecessary story about the episode. How could
we ever know that this is the way to view what occurs? How could we ever
tell that Lavoisier's new beliefs match nature more closely than his old ones
(the beliefs Priestley stubbornly retains)?
Kuhn's suggested economy is false economy. One way to recognize the
work that the idea of a fixed nature does is to see how we are impoverished
by giving it up. One difficulty is that it becomes troublesome to understand
the improvements in "concrete predictions" and "puzzle solving." We believe
that such improvements occur, for example in the transition from young
Lavoisier to mature Lavoisier, because we recognize that there is increased
harmony within our cognitive system: PI beliefs that are now induced in
scientists can be accommodated more easily within the total set of beliefs.
Realism provides a picture of the genesis of perceptual belief that attributes
a causal role to something beyond our cognitive systems. That picture has
the advantages that it offers an explanation of the apparent fact that some
of our beliefs come to us unbidden, it allows for the contents of our perceptual
beliefs to be partly determined by our prior cognitive state, and it enables us
to understand our seeming ability to achieve greater cognitive harmony in
terms of increased match between our representations and an independent
reality. To demonstrate that the "hypothetical fixed nature" is unnecessary,
antirealists must provide a rival picture that has similar virtues.8
8. The realist response does not claim to issue any guarantees here. There is no com-
pleting the Cartesian project of suspending all beliefs except those that are absolutely
certain and then demonstrating on this basis that the realist account of nature and our
relation to it is correct. Instead, realists counter with a challenge: what rival account will
explain how we are able to achieve greater cognitive harmony in coping with inputs that
Realism and Scientific Progress 133
3. Pitfalls of Pessimism
Contemporary biologists agree that human beings and chimpanzees had a
common ancestor no more than a few million years ago. I claim that their
assertion is true. Critics object. They demand to know why I think that what
the biologists say is true. How should they be answered?
I shall start by distinguishing two forms of antirealism. Global critiques
of realist epistemology contend that we have no basis for claiming that any
statement is true (when truth is understood in the realist's preferred way).
we are apparently unable to control? For a more detailed articulation of this general line,
see (Quine 1960).
9. This is not so much an explicit thesis in the writings of historians and sociologists as
a pervasive attitudeindicated in eschewing talk of truth or reality or in uses of scare
quotes. I have often heard it said in discussions that Kuhn has "disposed of notions like
truth" or has shown that "there is no theory-independent world." Because such claims are
so popular, any defense of a realist picture of science has to begin by showing that the
notion of truth makes sense and that it is a mistake to slide from the theory dependence
of our representations of the world to the theory dependence of the world. This is as far
as I have hoped to go in the present section.
134 The Advancement of Science
So the global antirealist would be skeptical about the truth of the biologists'
claim as a consequence of a sweeping generalization. By contrast, local cri-
tiques of realist epistemology allow that we are entitled to maintain that some
statements are true (in the realist sense) but they complain that statements
which realists take to be true, statements with some special feature U, outrun
this entitlement. So, one kind of local antirealist would deny our justification
for accepting as true statements about entities in the remote past and would
base skepticism about the biologists' thesis on this denial. The most famous
form of local antirealismone that has flourished in our centuryidentifies
the special feature U as that of making reference to unobservables.10 I shall
argue against this version of antirealism later, but I shall also suggest that
there are more congenial versions, indeed versions that give my account of
progress everything that it needs.11
Imagine, then, that we face an antirealist who doubts that what the bi-
ologists say about our ancestry is true. An obvious first response is to elaborate
the evidence that has convinced people about the recency of human-chimp
common ancestry. So one can explain the general idea of a molecular clock,
show how molecular clocks can be calibrated and how they give an estimate
of about five million years for separation of the human and the chimpanzee
lineages, and, as the coup de grace, rehearse the details of the fossil record.
This will not satisfy a global antirealist. For beyond the issue of demarcating
statements like this, which are accepted, from others (such as, Human beings
were directly created along with all other animal kinds about 10,000 years
ago) which are rejected, global antirealism's prime concern is how any state-
ment might earn the title of truth, when truth is conceived as in the realist
picture.
But there is an answer to this question. We have scientific views about
the relations between ourselves and the rest of nature. In the light of these
scientific views we can evaluate the likelihood that we are right about various
kinds of things. So we examine the procedures that are involved in accepting
the hypothesis about human-chimp separation and show, by appealing to
background ideas about reliable detection of aspects of nature, that such
procedures would regularly deliver truth (understood in the realist way).
10. Classical instrumentalism does not quite take this form, since, for the instrumen-
talist, it is not merely that we cannot justifiably assign truth values to statements purporting
to be about unobservables but this epistemic remoteness deprives the statements of mean-
ingfulness. The local antirealism that I characterize in the text is closer to the views of Bas
van Fraassen (1980), which are discussed later, and, perhaps, to those of the many nine-
teenth century scientists who worried about atoms, electrons, and the ether.
11. There are also global critiques of the type of realism I favor which might allow my
account of progress everything it needs, for example Arthur Fine's "natural ontological
attitude" (NOA) (Fine 1986). I must confess to finding NOA elusive: in his attacks on
realism, Fine seems to become an antirealist, and in his rejection of antirealism, he appears
to become a realist. Plainly this assessment is at odds with NOA's professed goal of slipping
between two antagonistic positions. If it can indeed find its way, then, perhaps, it can
accommodate large parts of my account of scientific progress (assuming, of course, that
those survive the criticisms considered later in this chapter).
Realism and Scientific Progress 135
Other procedures, such as those underlying the creationist claims that are
rejected, would not.
Thoughtful antirealists will not be entirely convinced. A first worry will
focus on the appearance of circularity. The envisaged defense of the truth of
the claim about human-chimp separation requires acceptance of certain parts
of contemporary science. Someone suspicious about molecular clocks or about
the fossil findings will not be persuaded. Nor, by the same token, can we
expect to satisfy those who think that our current physical-physiological-
psychological conceptions of the relationship between human cognitive sys-
tems and the environment are mistaken. Skeptics who insist that we begin
from no assumptions are inviting us to play a mug's game. Descartes's lack
of success in generating an account of nature that would survive all possible
doubt was in no way the result of deficiencies of intellect or imagination.
The global antirealist aims to challenge the status that is conferred on
accepted science as a whole. To respond it is necessary to show how our
scientific conception of the physical-physiological-psychological relationships
between human cognitive systems and nature supports theses to the effect
that certain types of processes and procedures yield true beliefsrepresen-
tations that match natureand to demonstrate that it is processes and pro-
cedures of these privileged kinds that are involved in the generation of
accepted science. By hypothesis the skeptic accepts our current science, and
that will include the conception of the relations between our cognitive systems
and nature.12 Hence the strategy seems entirely appropriate, for it starts with
premises that are accepted by those we intend to address. Moreover, even
though science is being used to defend itself, there is no guarantee of success
in our project. Application of our conception of the relation between human
cognitive systems and nature might convince us that some procedures crucial
to various parts of accepted science are unreliable: suppose, for example, that
we were able to show that large parts of contemporary genetics rest upon
experiments in which people are required to discriminate mutants in ways
that can be shown to outrun the abilities manifested in the test situations of
perceptual psychology. As we shall see, considerations of this type generate
versions of antirealism.
12. Nothing very grand needs to be involved in the early stages of this venture. Our
dialogue with the global antirealist might begin with our outlining some findings about the
performance of subjects in standard physiological and optical conditions when they are
confronted with objects of various shapes and colors. If the global antirealist endorses
claims about the regular covariance of belief with the stimulus objects, we can use this as
an entry to claims that we have ways of attaining truth, conceived realistically. If not, then
the global antirealist must dissent from so much of our contemporary scientific picture of
the world that we have an insufficient basis for responding to his skeptical challenge. Here
all there is to be said is that if the set of premises that he will allow is so curtailed, we have
no chance of generating the conclusion he demands: this failure should be no more surprising
than Descartes's inability to issue a complete guarantee for the truth of all those claims
that Meditation I calls into doubt. A priori guarantees are not to be had, and if the antirealist
refuses to allow us to appeal to relatively uncontroversial a posteriori premises, then the
task that he sets is impossible, and this should simply be admitted.
136 The Advancement of Science
Clever antirealists see that the issue is whether all or some of our accepted
beliefs have a particular status, and they use the fact of scientific change to
play reason against itself. Their best strategy is the so-called pessimistic in-
duction on the history of science.13 Here one surveys the discarded theories
of the past; points out that these were once accepted on the basis of the same
kinds of evidence that we now employ to support our own accepted theories,
notes that those theories are, nevertheless, now regarded as false; and con-
cludes that our own accepted theories are very probably false.14 It should be
apparent from my continued emphasis on practices rather than on theories
that I shall want to reformulate this pessimistic induction before replying to
it. But, even before doing so, I want to suggest that the history of science
provides grounds for optimism as well as pessimism.
There are a number of ways to present an optimistic view of the history
of science. One (offered in Devitt 1984 146) is to suggest that we look at the
track record of successful uses of theoretical terms. Perhaps most of the posits
introduced by our remote predecessors are entities we no longer countenance,
but, as time goes on, we find that more and more of the posits of theoretical
science endure within contemporary science. We attribute this increased suc-
cess to improved abilities of scientists to learn about investigating the world.
Hence, as Michael Devitt points out, the pessimistic observation that a lot
of past theoretical science has been discarded is not enough. The skeptic
needs to show that "the history of unobservable posits has been thoroughly
erratic" (Devitt 1984 147).
Although this line of argument is suggestive, and although it makes the
important point that confidence in the posits of contemporary science could
be justified, even if our first efforts were discredited, provided that our practice
of positing improved with time, it is vulnerable to some important challenges.
Antirealists will deny that they have to show that "the history of unobservable
posits has been thoroughly erratic," noting that confidence would be under-
mined if there were a relatively constant tendency, at all epochs in the history
of science, to introduce posits which, roughly as often as not, had to be
subsequently abandoned. They will also point out that if endurance simply
amounts to temporal persistence, then there were many entities introduced
within ancient scienceepicycles, humors, and so forththat survived for a
very long time before being discarded. To develop Devitt's argument, one
needs to capture the idea that the stability of modern references to molecules,
genes, and extinct organisms throughout periods of great theoretical prolif-
eration somehow redounds more to their credit than the persistence of
Aristotelian and Galenic conceptions through the Middle Ages. Finally, anti-
13. Sometimes known as "the disastrous meta-induction" (see Putnam 1978). I shall
examine the most sophisticated version of the strategydue to Larry Laudanin the next
section.
14. At this point there appear to be two options. One local antirealist conclusion is to
claim that this shows only that our claims about unobservables are suspect. Global anti-
realists, perhaps impressed by the theory-ladenness of all our beliefs, draw the more am-
bitious conclusion that all our current claims about nature are dubious.
Realism and Scientific Progress 137
claims and show that they do not invalidate reformed realism's contentions
that our theoretical statements are improving, our schemata becoming more
complete, our concepts more adequate, when all these are understood in the
realist's preferred way. So the pessimistic induction has no force against a
properly formulated realism." The optimistic induction, then, is not an at-
tempt to show that the history of science must be interpreted in terms of the
realist's conception of progress (although it is worth asking the antirealist to
explain a rival account of progress in detail and to show that it does not make
surreptitious appeals to truth) but to blunt an apparently damaging challenge
to explain the difference between the science of the past and the science of
the present.
Suggestive as they are, these observations strike me as altogether too
weak. The pessimistic induction relies critically on the gross analysis of science
as a constellation of theories which are supposed to be true and which are,
historically, failures. I oppose to it a simple point. The history of science does
not reveal to us that we are fallible in some undifferentiated way. Some kinds
of claims endure, other kinds are likely to be discarded as inaccurate. Fur-
thermore this is exactly what we would have expected given our conception of
the relationship between human cognitive systems and nature. According to
that conception we are relatively good at finding out some things, and discover
others only with relative difficulty. We would expect that, in the latter sorts
of endeavors, history would be peppered with false starts and plausible, but
misleading, conclusions. With respect to the former projects we would an-
ticipate doing better.
Instead of a blanket pronouncement to the effect that our current theories
are probably wrong, it would be far more instructive to investigate the stability
of various components of practice in various fields. Perhaps this investigation
would pose a challenge to parts of contemporary science, by revealing that
processes that we take to be reliable and that are critically involved in sup-
porting those parts of contemporary science have, in the past, given rise to
large numbers of erroneous or inadequate ideas. But this needs to be shown.
I shall argue later that the usual stock of examples which is supposed to
embarrass realists only issues in a form of antirealism that leaves my account
of progress untouched.16
Before I turn to this project, let me list some of the questions that would
have to be addressed in a more sophisticated study of the bearing of the
historical record on the status of current science. (1) How stable are the
reference potentials of scientific concepts? To what extent are refinements of
reference potential cumulative? (2) How frequently do scientists discard
16. In effect, the investigation that I recommend here turns the antirealist use of the
pessimistic induction on its head. Instead of starting with a preconceived thesis about the
feature U that distinguishes those statements that outrun our entitlement to claim truth,
we might use the history of science as a partial source for discovering our limits. Appeals
to history of science would be used in conjunction with studies of the cognitive performance
of subjects under natural and artitifical conditions to find out where we are most likely to
make mistakes and where we are legitimately confident.
Realism and Scientific Progress 139
17. It is important to recognize that while part of this picture is extremely recent and,
in some respects, controversial, other aspects of our views about cognition are well en-
trenched and have survived millennia. They are implicit in our everyday practices of ad-
justing our movements so as to enter into causal contact with things about which we want
to know. When I want to find out whether the plants on the patio are too dry I go outside
and look at them, sometimes pushing my fingers into the soil. Doubtless millions of other
people have done very similar things for thousands of years, and their behavior signals a
tacit conception of the relationship between their cognitive states and the world.
140 The Advancement of Science
reconceptualization that occurs when reference potentials are refined). Op-
timism about our current instruments and techniques is fostered if we discover
that they stand at the end of a sequence of such techniques, whose initial
members had high rates of success and whose subsequent members have
exhibited successively higher success. For that discovery fosters the idea that
we have learned about nature and, as Devitt suggests, learned ever more
about learning about nature.
Further historical research could cast a pall over our optimism. We might
find out, for example, that long sequences of apparent successes were punc-
tuated by embarrassing reversals, prompted by recognition that what we took
to be observational/instrumental/technical improvements were based on quite
faulty understanding of the relationship between ourselves and nature, so
that an instrument or experimental technique had misled us across the board.
If this were to occur frequently in the history of science it would challenge
our complacency about current instruments and techniques by raising the
probability that there are unanticipated sources of error even for long se-
quences of high and improving success rates. Yet, as I have indicated before,
the pessimist has to do considerable historical work to show that this is the
case.
Even more work is required to demonstrate that the use of unaided ob-
servation to arrive at claims about the macroscopic properties of medium-
sized objects is unreliable. For our descriptive lore about substances and
organisms seems remarkably stable, enduring from antiquity into the present
with, occasionally, episodes of reconceptualization. Given the account of
conceptual change developed in the past chapter, we can thus defend against
the suggestion that history reveals that we should not claim truth for our
ordinary judgments about ordinary things.
Quite plainly we are more likely to be wrong when we advance general
conclusions, when we make claims about things that are causally remote from
us, and when we attempt precise specifications of magnitude. Our scientific
picture of nature provides us with ideas of the limits of our accuracy (at least
in some cases). The world may be such that our ability to achieve true gen-
eralizations about it may vary significantly from field to fieldand we may
use what knowledge we do acquire to differentiate those instances in which
pessimism is warranted from those in which we are, justifiably, optimistic.
4. History Revisited
Global antirealism is not usually adopted as a direct consequence of a pes-
simistic induction on the track record of scientific theorizing.18 Instead of
18. Nonetheless, global antirealism is not without supporters. There are many passages
in his (1962/70) that seem to commit Kuhn to global antirealism (see the passages cited
previously, and almost all references to the idea that scientists separated by a revolution
live in "different worlds"). I take these to stem in large part from the worries about the
coherence of realist notions canvassed in Section 1. Many sociologists of science are plainly
Realism and Scientific Progress 141
convinced of the barrenness of talking about truthsee, for example, Latour and Woolgar
(1979), Shapin and Schaffer (1985), Collins (1985). Their global antirealism may simply be
based on Kuhn's influence, or it may result from the idea that a pessimistic induction shows
at least some version of local antirealism and that the recognition of theory-ladenness entails
that no distinctions can be drawn among statements.
19. Although it is clear that Laudan does not question the claim that science makes
progress, I am not certain whether he should be counted as a local antirealist who takes
realism to overstep its bounds in defending the truth of "high theory" (a local antirealist
who takes U to be the property of being a "theoretical" statement). If Laudan were to
resist realist claims about the truth of nontheoretical statements, then he would seem to
assume the burdens of global antirealism mentioned in the last section. Indeed, his argu-
ments from the history of science do not seem to be at all directed toward the sweeping
conclusions that a global antirealist would have to establish. Furthermore, accepting the
idea of realist truth for nontheoretical statements would enable him to understand the idea
of a puzzle solution in an obvious way, namely as the production of true, nontheoretical,
answers to questions. The cost, quite evidently, would be to embrace a distinction very
like that once drawn between observational and theoretical statements, and, since Laudan
accepts the Kuhn-Feyerabend-Hanson critique of theory-independent observation, this
would create considerable tension within his position. That tension is muted in practice
because of his reticence in specifying the conditions that are required of a puzzle solution.
In what follows, I shall ignore difficulties of interpretation, focusing on whether Laudan's
historical examples serve to show that the kinds of statements with which he is centrally
concerned fall beyond the range of realist entitlements.
142 The Advancement of Science
Not so, I claim. Laudan's argument depends on painting with a very broad
brush. We are to think of "theories," "success," and "referential central
terms."20 This discourse yields at best only spurious correlations.
To diagnose part of the trouble, let us look at an extension of Laudan's
argument, intended to block the charge that matters have altered in the
twentieth century.
Consider, for instance, virtually all those geological theories prior to the
1960's which denied any lateral motion to the continents. Such theories were,
by any standard, highly successful (and apparently referential); but would
anyone today be prepared to say that their constituent theoretical claims
committed as they were to laterally stable continentsare almost true? (1984
123)
There are two points that ought to be conceded to Laudan at the outset.
First, there were indeed large parts of early twentieth century geology that
offered successful explanations and predictions of various phenomena, and
which survive, sometimes reconceptualized, in current geological practice.
Second, although there was, from 1915 on, a minority tradition that asserted
the lateral motion of continents, the individual practices of most influential
geologists, especially in the United States between the wars, contained the
statement that the continents do not move laterally. Is this enough to support
Laudan's antirealist conclusion? I do not think so. For the point of the ex-
cursion through the history of science is to convince us that the apparent
successes of contemporary science might be misleading, that the terms of the
languages of currently successful practices do not refer. To produce conviction
on this point Laudan must establish more than the points I have conceded.
It will be necessary to show that the denial of the lateral motion of continents
actually plays some role in the successes of the old practice. No sensible realist
should ever want to assert that the idle parts of an individual practice, past
or present, are justified by the success of the whole.21
20. In Laudan's defense it might be noted that this is the vocabulary used by the
proponents of realism whose claims he is out to discredit (see, for example, Putnam 1975).
My own realist conception of progress builds on the framework introduced in Chapter 3,
and, I shall argue, provides the means to distinguish what is correct in Laudan's historical
claims from his antirealist conclusions in ways that were unavailable to the realists whom
he attacks.
21. This point is akin to one of Hempel's great insights about the logic of confirmation:
one cannot generate an acceptable account of confirmation by supposing both that (a) if e
is a consequence of h then e confirms h, and (b) if h' is a consequence of h then any
statement that confirms h confirms h'. Joint acceptance of these conditions yields the
conclusion that any e confirms any h. For (by (a)) e confirms e & h; h is a consequence of
e & h; so (by (b)) e confirms h. (See Hempel 1965, chapter 1.) One obvious moral is that
confirmation does not accrue to irrelevant bits of doctrine that are not put to work in
delivering explanations or predictions. Since sensible realists ought to accept this point,
the extra demand made in the text is well motivated. As Laudan recognizes, at least one
realist (Glymour 1980), perhaps the realist most attuned to Hempel's insights about con-
firmation, rejects the hyperextended holism that spreads success over all accepted sentences
without asking which have been put to work.
Realism and Scientific Progress 143
22. Laudan (1984 116-117) offers some remarks that suggest that he can accommodate
this kind of argument. He contends that any "weakening" of realist claims to allow that
some "components" of successful past theories were false/nonreferential will prove disas-
trous for realism. But this is incorrect if the realist can distinguish between those parts of
theory that are genuinely used in the successes and those that are idle wheels. Without
that kind of distinction everyone's account of confirmation is going to be in trouble for the
generic reasons presented in footnote 21. Moreover, I suggest that the move from thinking
of "theories" to considering practices, and, specifically, to focusing on the set of schemata,
will enable us to see how to make this distinction.
23. The geological case just considered is atypical of Laudan's inventory of examples,
in that here Laudan does not make any claim about failures of reference of central terms.
I had originally overlooked this difference, and I am grateful to Gary Hardcastle for
convincing me that I was wrong.
144 The Advancement of Science
were a number of accepted statements, well deployed in accepted schemata,
which seemed to presuppose the existence of the ether. A number of eminent
individual scientists also advanced claims about the nature of the ether, some
of which secured wide acceptance ("The ether is an elastic solid"), others of
which were controversial (the mathematical descriptions offered by Augustin-
Louis Cauchy, George Green, and James MacCullagh). Before we can assess
Laudan's thesis it is important to separate different enterprises within the
physics of the period.
Laudan's central concern seems to be with the most obviously successful
of these ventures
Within the theory of light, the optical aether functioned centrally in expla-
nations of reflection, refraction, interference, double refraction, diffraction,
and polarization. (Of more than passing interest, optical aether theories had
also made some very startling predictions. E.g., Fresnel's prediction of a
bright spot at the center of the shadow of a circular disc was a surprising
prediction that, when tested, proved correct. If that does not count as em-
pirical success, nothing does.) (1984 114)
We need to ask two questions about this situation: first, what statements of
the theory were deployed in providing these explanations and predictions?
second, do these statements contain nonreferring termsin particular, do
they involve putative references to ethers or subtle fluids?
Consider, in this context, Fresnel's prediction of the Poisson bright spot.24
On a superficial reading, Fresnel's derivation of descriptions of the fringes
produced in diffraction is reassuringly familiar. He appeals to Christian Huy-
gens's conception of the wavefront as a center of secondary wave propagation,
applies the principle of interference, and computes the resultant wave motions
at the screen (or the eye) using the mathematics that physics students learn
from textbooks on optics. First, Fresnel shows how to resolve a single wave
into two component waves that differ in phase by Y/2. He then calculates the
wave motion at a point on the screen produced by an infinitesimal arc segment
of the wavefront. Integration yields the entire contribution of the wavefront,
and the resultant formula enables Fresnel to plot the expected distribution
of illuminated and unilluminated zones (and, more generally, the intermediate
degrees of partial illumination).25
24. Strictly speaking this consequence was not derived by Fresnel. It is a consequence
of the general treatment of diffraction proposed in Fresnel's prizewinning memoir of 1818,
published in 1826. (See Fresnel 1856/1965 247-382.) In the first note to this memoir, Fresnel
explains, "M. Poisson communicated to [him] the singular result that the center of the
shadow of a circular disc should be illuminated as if the disc were not there" (Fresnel 1856/
1965 365). Although Arago and Fresnel himself subsequently confirmed this striking con-
sequence, Poisson was never convinced (see Worrall 1989).
25. Geoffrey Cantor (1983) gives an accurate digest of Fresnel's method in the case of
occlusion by a semiinfinite body (152-155). For Fresnel's own discussion, see the 1818
memoir in (Fresnel 1865/1956), especially 286-293, 313-316. Compare the contemporary
treatments offered, at a relatively elementary level but with extraordinary lucidity, by
Richard Feynman et al. (1963 I 29.1-29.7, 30.1-30.3), and, with rigorous mathematical
detail by Max Born and Emil Wolf (1980 370-375).
Realism and Scientific Progress 145
Descartes, Hooke, Huyghens, Euler, thought that light resulted from the
vibrations of an extremely subtle universal fluid, set in motion (agite) by the
rapid movements of the particles of luminous bodies, in the same way that
the air is stirred (ebranle) by the vibrations of sounding bodies . . . (1818, in
1865/1965 248)26
26. Here and elsewhere, translations from Fresnel are my own. In the paragraphs
that follow, I shall simply give page references to Fresnel's memoir as reprinted in (1856/
1965 I).
146 The Advancement of Science
Moreover, the ether makes a second entrance when Fresnel discusses the
interference of light waves. First, his statement of the problem to be solved:
Given the intensities and the relative positions of any number of systems of
light waves of the same wavelength, which propagate in the same direction,
to determine the intensity of the vibrations resulting from the joint action
(concours) of these different systems, that is to say the velocity of the os-
cillations of the molecules of the ether (la vitesse oscillatoire des molecules
etherees). (286)
HR. Fresnel's dominant intention is to talk about light, and the wavelike
propagation of light, however that is constituted. He has, of course, a
false belief about the medium of propagation. But, since his primary aim
is to discuss light and its wavelike qualities, his tokens of 'light wave' in
the solutions of the diffraction and intereference problems genuinely refer
to electromagnetic waves of high frequency.
HN. Fresnel's references are explicitly fixed through the descriptions
he gives at the beginning of the memoir in characterizing the wave theory
of light and in introducing the wave equation. Tokens of 'light wave' that
occur later in the memoirfor example in the solutions to the problems
of diffractionthus have their references fixed by the description "the
oscillations of the molecules of the ether." Since there is no ether, they
fail to refer.
27. See the illuminating discussion by Cantor in his (1983), especially 158, and also Jed
Buchwald's (1981) and (1989, especially 306-310). Buchwald is primarily concerned with
Fresnel's treatment of polarization rather than with his discussions of interference and
diffraction.
Realism and Scientific Progress 147
the ether. Those should be given their due but should not be seen as some sort of creeping
rot that invades everywhere. Laudan's penchant for tarring past theories as a whole with
local misconceptions seems to me to be as unbalanced as neglecting every aspect of the
past that does not fit within contemporary views of nature. Moreover, it signally fails to
explain how the textbook presentations of historical figuresthe whig history of the science
textis ever possible in the first place.
31. Kenneth Schaffner adeptly chronicles the struggles in his (1972). There were oc-
casional successeswitness Fresnel's "ether drag" hypothesis and the solution of the prob-
lem of aberration (see footnote 28)but the overall picture is one of highly sophisticated
theorists proving that yesterday's account of the ether won't do (because it is dynamically
unstable, unable to yield some one of the properties required of the ether, and so forth).
Interestingly, there were always doubters, such as George Airy, impressed with Fresnel's
mathematical account of wave propagation but agnostic about the medium in which the
waves were propagated.
Realism and Scientific Progress 149
erential status of those terms that are frequently used in successful schemata,
we do learn something important from the example. Distinguish two kinds
of posits introduced within scientific practice, working posits (the putative
referents of terms that occur in problem-solving schemata) and presupposi-
tional posits (those entities that apparently have to exist if the instances of
the schemata are to be true). The ether is a prime example of a presuppo-
sitional posit, rarely employed in explanation or prediction, never subjected
to empirical measurement (until, late in the century A. A. Michelson devised
his famous experiment to measure the velocity of the earth relative to the
ether), yet seemingly required to exist if the claims about electromagnetic
and light waves were to be true. The moral of Laudan's story is not that
theoretical positing in general is untrustworthy, but that presuppositional
posits are suspect.
Laudan recalls, with considerable rhetorical effect, James Clerk Max-
well's remark that "the aether was better confirmed than any other theo-
retical entity in natural philosophy" (Laudan 1984 114; the formulation is
Laudan's not Maxwell's). Although we can understand his claim, based as
it was on the multiplicity of phenomena to which schemata appealing to
wave propagation had been successfully applied, Maxwell was wrong. The
entire confirmation of the existence of the ether rested on a series of paths,
each sharing a common link. The success of the optical and electromagnetic
schemata, employing the mathematical account of wave propagation begun
by Fresnel and extended by his successors (including, of course, Maxwell),
gave scientists good reason for believing that electromagnetic waves were
propagated according to Maxwell's equations. From that conclusion they
could derive the existence of the etherbut only by supposing in every
case that wave propagation requires a medium. Thus the confirmation of
the existence of the ether was no better than the evidence for that
supposition.
Contrast the ether with the working posits of theoretical science that are
referred to and characterized directly in successful schemata: atoms, mole-
cules, genes, electromagnetic fields, and so forth. In some cases, we even
devise numerous techniques for measuring the magnitudes of quantities that
these entities possess or for representing the entities in various ways. If Lau-
dan's story has an antirealist moral, it is that the presuppositional posits of
contemporary science may not exist.
Laudan's use of the pessimistic induction from the history of science to
discredit the claims to reference and truth of current science and to undermine
a realist account of progress is the most sophisticated that I know. I have
been arguing that it is not sophisticated enough. It leaves untouched the
central realist claim. A finer-grained look at the history of science shows that
where we are successful our references and our claims tend to survive even
extensive changes in practice and to be built upon by later scientists, giving
us grounds for optimism that our successful schemata employ terms that
genuinely refer, claims that are (at least approximately) true, and offer views
about dependencies in nature that are correct.
150 The Advancement of Science
32. In this, my approach is akin to the positions of Hilary Putnam and Richard Boyd,
whom van Fraassen cites, who take realism to involve claims about actual attainments.
See, for example, (Putnam 1975 69-75).
33. Van Fraassen thinks that the "naive statement" of realismwhich supposes that
"today's theories are correct"lumbers the realist with something too strong (1981 7-8).
But the trouble here isn't with claims about correctness. Rather it lies with the attribution
of truth to whole theories. Where van Fraassen qualifies the naive statement by retreating
from commitments to achievements, I make it more cautious by specifying the attainments
more narrowly. See the discussion in Section 6 of the last chapter.
Realism and Scientific Progress 151
understand human cognitive limitations will see that the whole truth is not
something that we want and that, in some cases, the attempt to achieve truth
(rather than approximation or idealization) would interfere with our primary
aims.
The debate between van Fraassen and his realist opponents is an internal
dispute within the broader controversy about realism discussed in Section 2.
Van Fraassen is willing to allow the possibility of arriving at some kinds of
true beliefs. His opposition to realism is based on maintaining that the domain
over which we can reliably generate true beliefs is limited to observables: van
Fraassen is a local antirealist for whom the feature U that marks the limit of
our defensible claims is the property of referring to unobservable entities.
He suggests that the aim of science should be to achieve theories that are
empirically adequate, that deliver the truth about observables.
Once again, I think that van Fraassen's formulation skews the issue in an
important way. Let us ask which truths about observables we would want an
ideal science to deliver. All of them? Surely notfor that would bury what
we really want to know in a clutter from which our limited cognitive systems
could not retrieve the desired information. The significant ones? But, in this
case, what does significance amount to? It is far from clear that van Fraassen's
framework allows him to offer the type of account of cognitive significance
that I sketched in the previous chapter.
I shall not pursue this line of inquiry, because I think that there is a
fundamental difficulty with van Fraassen's project of limiting science to the
observable. Why should we settle for less than the delineation of natural
kinds, the provision of correct schemata, the answering of significant ques-
tions? Van Fraassen's objections to these apparently desirable but ambitious
goals would seem to be that they are too ambitious. Nothing in his writings
suggests that it would not be a good thing to know about the behavior of
unobservables. The line of argument appears to be that these goals are beyond
our epistemic reach.
Given the discussion of the pessimistic induction in the previous section,
this is a species of argument we should recognize as exactly to the point. My
general account of scientific progress takes the following form: we would like
our components of practice to achieve the qualities X, Y, Z, and so forth; if
our account of ourselves shows that we are limited in attaining one of these
qualities in certain cases, then we should seek approximations or idealizations
that advance the overall project of fathoming the structure of nature. If van
Fraassen were able to show that our cognitive limitations prevent us from
any knowledge of the unobservable, then the idea of scientific progress that
I have delineated would be unfeasible and some modification of it in the
direction of the antirealist aim he advances would be motivated. But if there
is no reason to believe that our cognitive limitations are so great that they
bar knowledge of the unobservable, then there is no reason to abandon the
account of progress I have outlined. Van Fraassen's argument is of the right
sort because considerations of human cognitive limitations should be used in
formulating goals that are attainable for us. Whether it is a cogent critique
152 The Advancement of Science
34. Sec many of the essays collected in (Churchland and Hooker 1984), particularly
those by Paul Churchland, Ronald Giere, Gary Gutting, and Alan Musgrave, and also
(Railton 1989).
Realism and Scientific Progress 153
successful organisms are those that leave descendants, but also to investigate
those characteristics that promote reproductive success, carrying out the in-
quiry at whatever level of generality is possible. Thus we can distinguish (i)
current species and those that have endured for significant periods in the past;
(ii) the possession of high Darwinian fitness, borne by those organisms that
have characteristics that dispose them to survive and to reproduce most suc-
cessfully; (iii) the generic characteristics that endow organisms with high
Darwinian fitness (e.g., ability to avoid predators); and, finally, (iv) the par-
ticular ways in which organisms belonging to particular species achieve these
generic characteristics. Similarly, in the case of scientific theories, we should
separate (i) current theories and those that have endured for significant pe-
riods in the past; (ii) the possession of high propensities for espousal by
scientists, borne by those theories that have great predictive and explanatory
power; (iii) the generic characteristics that endow theories with great predic-
tive and explanatory power; and, finally, (iv) the particular ways in which
particular theories achieve these generic characteristics. Van Fraassen sug-
gests that the presence of certain species is to be explained by noting that the
organisms belonging to them have high fitness in the relevant environments
and this is indeed a relatively shallow level of Darwinian explanation, cor-
responding to the understanding of (i) in terms of (ii).39 By analogy, we could
explain the endurance and current presence of theories which appear suc-
cessful by remarking that those which did not have high explanatory and
predictive power tended to be discardeda relatively shallow level of expla-
nation that again explains (i) in terms of (ii).
This should not be the end of the story. Darwinians want to know, at
whatever level of generality is possible, what kinds of organism-environment
relationships confer reproductive successthey want to explain (i) and (ii)
by moving on to (iii). Similarly, Putnam's demand is to ask what it is about
those theories that are explanatorily and predictively successful and that posit
unobservables that makes for their explanatory and predictive success. The
answer given by the realist is that such theories succeed in this way because
they fasten on aspects of reality. If they did not, it would be "a miracle" that
they were so successful. Now that answer can be challengedas it is chal-
lenged by the pessimistic inductivistsby claiming that highly successful the-
ories frequently introduced posits that we no longer recognize. But, as we
saw in the previous section, that challenge can be turned back. I thus suggest
that a parallel to the thoroughly Darwinian strategy of trying to elicit the
general characteristics of the organism-environment relation that confers
6. Goals in Flux?
40. This position is elaborated in his (1984), and there are some tensions between this
later work and the view of progress in terms of puzzle solving advanced in (Laudan 1977).
I shall not explore possible ways of resolving the tension, since I am primarily interested
in the challenge posed by Laudan's claims about variable aims.
158 The Advancement of Science
where, in each case, L L' means that the change is rational. Laudan does
not say a great deal that enables us to fill in the blanks in these statements,
but it is clear that, in general, the rationality of changing one L practice to
another involves remedying various kinds of inconsistencies among elements.
In the special case of changes of aims, corresponding to my third schema, he
does propose that we revise aims in the light of theories and methods so as
to devise substitutes for goals that come to be regarded as unattainable.42
There are several worrying features of this reticulational model: for in-
stance, the apparent difficulty of saying how, in a clash between theory and
methodology, we should decide whether to revise theory or to change meth-
odology, and the seeming possibility that the state of L practice at the end
41. Indeed, this is supposed to have occurred at various points in the history of science.
See the discussion of George Lesage's defense of the "method of hypothesis" and its impact
on Victorian views about the goals of science (1984 59).
42. See (Laudan 1984 51-53). There is a second constraint on the advocacy of a set of
goals, which receives more prominence in (Laudan 1989), namely the requirement to
"preserve the canon." Here the scientist is supposed to espouse a set of aims A that will
fit with the "recognized past achievements of science." It is interesting to ask whether this
demand is itself grounded in some enduring, context-independent goal for the enterprise
and, if not, why it should be imposed.
Realism and Scientific Progress 159
of a sequence of changes might look with disfavor upon some of the transitions
that generated that L practice (see Doppelt 1986). But the fundamental con-
cern that I have here is with the status of the principles that result when the
correct substitutions are made in the preceding schemata. What makes the
principles of rational transition among L practices the right principles?
Since it is clear that Laudan regards rational transitions among L practices
as relieving inconsistencies, and that he does not think that every way of
remedying inconsistency is rational, the demand is for understanding why
relief from inconsistency should be sought and why it should not be sought
in the obvious fashion of hacking away at our L practices until they are so
minimal as to avoid the contradictions. The obvious answer is that there are
enduring goals that are not represented in Laudan's official picture and that
the rules of rationality formulate good strategies for attaining these goals (see
the discussion of the next chapter). We eliminate inconsistency because we
want our claims to be true and we do not do so by means of some scorched
earth policy because we want as much significant truth as we can get.43 But
if Laudan were to offer this explanation, his thesis that the goals of science
vary across time would be reformulated so that it was no longer at odds with
the picture of progress I have offered. All that would vary would be the
formulation of derivative goals, and that type of variation is captured in my
framework by the conception of shifts in the notion of significance. On the
other hand, if Laudan does not explain the status of the principles of rational
L practice transition by appealing to their promotion of enduring goals, it is
unclear that he has any basis for supposing that these principles have any
special quality. Anyone is free to characterize a set of transitions with some
honorific term, but the significance of doing so needs to be explained.
The argument so far is intended to show that if there is to be rational
change in scientific axiology there have to be some enduring goals that provide
a basis against which judgments of rationality make sense. Once this conclu-
sion is reached, there is a natural explanation of apparent variation in the
goals of science. Scientists seek to present true answers to significant ques-
tions, insofar as they are able. As their views about objective dependencies
43. Laudan has suggested to me that empirical adequacy would do as well as truth in
grounding the demand to eliminate inconsistency with minimal mutilation. However, this
suggestion is both problematic and beside the point. First, the demand of empirical adequacy
will not require us to eliminate all inconsistencies. Assuming that there can be rival em-
pirically equivalent theories which make incompatible theoretical claims, the search for
empirical adequacy, rather than truth, will not require us to choose between them. Second,
even if it were true that empirical adequacy would underwrite the demand, this would not
obviate the need for some enduring aimspossibly disjunctiveto stand outside the se-
quence of L practices and ground the principles that govern rational adjustment. The point
here is fundamental. If the aims that are represented within the L practices are indeed
revisable, it cannot be in virtue of them that the principles of rationality that govern the
adjustment of the components of L practices obtain their force. This is fundamental because
once the commitment to some enduring aims is established, I believe that the account of
those aims which I offered in the last chapter is both natural and defensible by the historical
record.
160 The Advancement of Science
change, what they take to be significant will shift. Moreover, in light of the
statements they accept, they will view certain kinds of questions as admitting
exact answers, others as approachable in terms of idealizations or approxi-
mations, and yet others, perhaps, as unanswerable by human beings. Changes
in the statements accepted will be reflected in these appraisals, leading to
differences in the formulation of derivative aims.
The explanation I have just sketched should be familiar from discussions
of ethical relativism.44 Apparent differences in moral judgments may stem
from differences in belief against a background of shared fundamental values.
This, I suggest, is exactly what occurs in the history of science. Laudan's
reticulational model is driven by the idea that there are genuine modifications
of fundamental aims of science from epoch to epoch. On closer view, I claim,
changes in formulations of the aims of science can be understood as expres-
sions of the enduring goal of discovering as much significant truth as human
beings can in the light of changing beliefs about what is significant, what
nature is like, and what the nature of our relation to nature is.
Laudan's principal example of axiological shifts in the history of science
concerns an alleged oscillation between demanding that theories be (at least
approximately) true and demanding merely that they "save the phenomena"
(1984 44-66, 1988). Despite the fact that he is an adept chronicler of disputes
between those who formulate these aimsparticularly in the context of eigh-
teenth century physical scienceLaudan misses a simple explanation of the
evidence he cites. It is hard to believe that any of the major protagonists in
these debates would have dissented from the goals that were identified in the
last chapter. All would have hoped to formulate explanatory schemata that
captured the dependencies of phenomena, to pose and to answer questions
picked out as significant by such schemata to the extent that the formulation
of the schemata and the answering of significant questions were possible for
human beings. The dispute is not about desirability but about attainability.
Those who settle for less, arguing that hypotheses should "save the phenom-
ena," do so not because they think it undesirable to achieve true (or ap-
proximately true) hypotheses but because they think it beyond the power of
the human scientists to discriminate the truth from the rival possibilities.
7. Social Interference
It should be apparent by now that my account of progress is not committed
to the idea that truth is easily won nor to the view that the facts "force
themselves upon us" if the conditions are right. Achieving the truth is tricky,
but not uniformly so. Moreover, the history of science cautions us against
writing off any domain of investigation as inaccessible. Given the finiteness
of our collective lifespan, the boundedness of our distribution in space, and
the limitations of our cognitive systems, it is highly likely that there are some
44. See Brandt (1959), and the brief discussion of the last chapter.
Realism and Scientific Progress 161
aspects of the universe that we shall never be able to fathom. However, it is
not clear that we can come to know which these are without an enormous
amount of effort. Casual attempts to write off certain phenomena as beyond
the reach of science are typically premature. After all, who would have
thought that we would have come to know as much as we do about the
composition of matter or the history of life? Recall how many eminent bi-
ologists before Darwin despaired of ever understanding the origins of species,
the "mystery of mysteries."
Contemporary work in the sociology of science embodies a vigorous cri-
tique of the confident attitude that I have adopted. It would be simple to
respond to that critique by pointing out that it assumes that views of science
which suppose that realist goals are set and attained inevitably embody a
naive epistemology on which the human mind is transparent to nature. Barry
Barnes presents the approach he hopes to undermine as follows:
Although it would be generally agreed that biasing factors occur, in different
degrees, in most institutional contexts, and hence that most belief systems
embody some degree of error and distortion, the natural sciences are gen-
erally accepted as true and undistorted bodies of knowledge; their methods
as impartial, unbiased models of investigation. Thus, science can function as
a model of how we would be able to orient ourselves to the world in the
absence of our psychological biases and social prejudices. (1974 5)
Similar suggestions that one can only maintain that science aims at and attains
truth by neglecting the complexities of the processes that give rise to belief
and adopting a simple-minded epistemology can be found in a justly famous
study of daily laboratory practice.
At the onset of stabilisation [roughly, the point at which a statement becomes
accepted within the relevant community: PK], the object was the virtual
image of the statement; subsequently, the statement becomes the mirror
image of the reality "out there." Thus the justification for the statement TRF
is Pyro-Glu-His-Pro-NH2 is simply that "TRF really is Pyro-Glu-His-Pro-
NH2." At the same time, the past becomes inverted. TRF has been there all
along, just waiting to be revealed for all to see. The history of its construction
is also transformed from this new vantage point: the process of construction
is turned into the pursuit of a single path which led inevitably to the "actual"
structure. (Latour and Woolgar 1979/1986 177)
The quick response is simply to slough off the epistemological ideas that these
writers attribute to their opposition and then to proceed with one's business.
The fundamental realist thesis is that we arrive at true statements about
the world. That thesis does not imply that we have unbiased access to nature,
merely that the biases are not so powerful that they prevent us from working
our way out of false belief. Nor should realists be seen as embracing a curious
kind of positivism: the justification for belief that p is to be p (a schema that
some epistemologists once held for foundational beliefs"I have a sensory
experience of red"but which has been universally abandoned and was never
162 The Advancement of Science
maintained for beliefs about the structure of molecules!).45 Intelligent realists
will claim that the justification for belief about the structure of TRF is a (very
complicated) process involving interactions between scientists and nature,
and that the belief is justified because processes of this general sort regularly
and reliably generate true beliefs.46
Showing that the sociological attack on realism works by foisting upon
realists naive epistemologies that they do not adopt and should not want is
quick but unsatisfying. There are deep and important insights in recent dis-
cussions in the sociology of science, and these are missed unless realists meet
the fundamental challenge. Is there an account of our relationship to the
realists' supposed "independent reality" that will both accord with the detailed
accounts we have of the social lives of scientists and support the claim that
we gain knowledge of this "independent reality"?47
According to the picture developed in Chapter 3, changes in consensus
practice result from modifications of individual practice, in accordance with
rules for consensus formation. Those changes in individual practice, in turn,
result from two types of process: conversations with peers and encounters
with nature.48 The encounters with nature involve the impinging of stimuli
on cognitive systems whose prior states play an important role in determining
the new state of the system. Thus there is no suggestion that experience is
raw or unbiased. Those prior states are partially caused by conversations with
peers, earlier encounters with nature, and, ultimately, the training process
and even the earlier intellectual ontogeny of the scientists. The deep point
of the sociological critique is that the social forces that operate in this mod-
ification of practicethe rules for consensus shaping, the conversations with
peers, the training process and broader socialization within a larger com-
munitymay be sufficiently powerful that the effects of nature are negligible.
If correct, this point would undermine my account of progress from within,
accepting the idea that we can use our scientific picture of nature and our
relation to it to appraise our beliefs about nature, and maintaining that when
the appraisal is done it reveals that relationship to be too tenuous to deliver
true beliefs.49 The persistence of our judgments about naturethe persistence
45. See (Schlick 1934/1959) and, for a review, (Chisholm 1966). Foundationalist theories
of knowledge, which supposed that certain statements are self-warranting, were always
controversial, and, more often than not, associated with antirealism (specifically phenom-
enalism) rather than realism (see Ayer 1952). For sustained critique, see (Sellars 1963
essays 3 and 5).
46. The locus classicus for this approach is (Goldman 1986). The general point about
the need for introducing appeals to the psychological causes of belief into accounts of
justification is made by Gilbert Harman (1971), Alvin Goldman (1979), Hilary Kornblith
(1980), and me (1983a chapter 1). For further discussion, see Chapter 6.
47. This is surely the point at which one should engage the arguments of Barnes and
of David Bloor, both of whom hold that there is an independent reality, but that our
interactions with it radically underdetermine the judgments we can make about it. See
(Barnes 1992) and Bloor (1974/1991 especially 31-40).
48. Here I oversimplify. Recall that this is not an exclusive and exhaustive dichotomy.
49. Chapter 8 will explore this potentially worrying predicament in considerable detail,
Realism and Scientific Progress 163
attempting to understand the dynamics of introduction and diffusion of ideas that might
challenge parts of consensus practice.
164 The Advancement of Science
can be modified as a result of causal interactions with nature and the modi-
fications can spread through the community to enter consensus practice.
I hope that it is now apparent that the sociological critique is both intel-
ligible and difficult to assess. The issue that it raises should recall a familiar
dispute. Organismal phenotypes are the result of an interaction between the
genotype and the organism's environment. Disentangling the relative contri-
butions of the two has proved highly problematic. Especially with respect to
those behavioral traits that most fascinate us (tendencies to sex and violence)
there has been a pronounced tendency to gravitate to one of two polar po-
sitions. Either it is proposed that the genes exert an iron grip (so that no
matter how the environment is varied the same phenotype will result) or it
is suggested that the genes provide no constraints at all (for any given genotype
we can produce the phenotype we want by varying the environment). As
soon as the extreme positions are articulated, they are quickly disavowed:
everybody agrees that there are two important determinants. But there is still
room for dispute about relative importance, about the extent to which vari-
ation in phenotype is limited by genotype.50
In the present case, too, there are two types of factor which prove hard
to disentangle. The practices of individuals respond both to stimuli from
external, asocial nature and to the remarks made by teachers, friends, col-
leagues, and adversaries. The consensus practice responds to changes in in-
dividual practices that are mediated by the social structure of the community.
The position that the sociologists attack is the extreme suggestion that social
determinants make no difference: given the same inputs from asocial nature
there will be the same modifications of consensus practice, no matter what
the social structure. They counterpose the extreme view that inputs from
nature are impotent, or at least impotent with respect to the framework of
consensus practice: given the same social structure, there will be the same
modifications of consensus practice, no matter what the inputs from asocial
nature, and, in particular, the framework of consensus practice will remain
unaltered. I claim that there is a vast, unexplored middle ground between
these extremes.
What does a realist account like mine need? It is not necessary to assume
the thesis of social transparency that traditional philosophy of science pre-
supposes and that sociologists rightly attack.51 Nor is it enough simply to deny
50. I have discussed these issues at some length in my (1985b). See also (Hanna 1985)
and (Sober 1988).
51. The thesis of social transparency is simply the claim that ultimately all the modi-
fications of individual practice (or, in traditional terms, changes in belief) produced by
interactions with other members of the community can be seen as directed by the expe-
riences that gave rise to cognitive changes in those who transmitted information (or in
people who transmitted information to them or . . .). Thus we can think of a cognitive
subject as a kind of epistemic octopus whose tentacles reach out through the lives of others
to obtain vicarious experience, and we can simply explore the modifications of cognitive
state in light of this extended experience. At the back of this picture seems to lurk the
notion of some primal state in which beliefs are simply induced by nature without any social
interactions. I think that the sociologists are correct to disavow any such primal state and
Realism and Scientific Progress 165
the sociologists' polar antithesis. The relevant issue is whether, given the
actual social structures present in scientific communities, the input from aso-
cial nature is sufficiently strong to keep consensus practice on track. (Similarly,
in discussions of gene-environment interaction, the appropriate questions
concern whether, given the genotypes actually present, modifications of en-
vironment can be sufficiently strong to produce modifications of phenotype.)
The issue can be framed precisely as follows: Let us suppose that there
is a community of scientists with a consensus practice Cl, a distribution of
individual practices, and a particular social structure. Let C2 be a practice
such that, given the way the world is, the shift from C1 to C2 would be
progressive in any of the senses articulated in the last chapter. The facilitating
encounters for the shift from Cl to C2 are those sequences of stimuli (if there
are any) that would lead the community from its initial state to a subsequent
state in which the consensus practice is C2. My account of progress does
depend on the idea that, with respect to the kinds of consensus practices,
distributions of individual practices, and social structures actually found in
science, there are facilitating encounters actually available for scientists to
make progressive shifts. Ideally, I would like to contend that this obtains for
every facet of scientific practice, so that there is no inviolable framework that
input from nature cannot displace. However, a conservative version of the
sociological critique would be compatible with the partial applicability of my
conception of cognitive progress.
So far I have merely tried to arrive at a clear view of what is at stake. Let
us now return to the claims and arguments of skeptical sociologists. Latour
and Woolgar appear to commit themselves to the extreme position that I
have delineated:
We do not use the notion of reality to account for the stabilisation of a
statement . . . because this reality is formed as a consequence of this
stabilisation.
We do not wish to say that facts do not exist nor that there is no such
thing as reality. In this simple sense our position is not relativist. Our point
is that "out-there-ness" is the consequence of scientific work rather than its
cause. (Latour and Woolgar 1979/1986 180-182)
to reject any analysis of the grounds of cognition that attempts to reduce the social inputs
to complexes of asocial interactions with nature.
166 The Advancement of Science
practices are all the same, then there would be powerful support for Latour
and Woolgar's thesis of social determination and the construction of facts. If
they vary, with the variation perfectly correlated with the substances pre-
sented, then there would be powerful support for the old-fashioned idea that,
at least for this kind of social arrangement and this kind of statement, input
from reality is strong enough to keep science on track. Other results would
testify to an interaction between the social factors and inputs from asocial
nature.
We cannot perform the experiment. But there is a common feeling that
we know in advance how it would turn out. Surely, we think, there would
be some variation: some of the groups would adopt a different belief about
the structure of TRF or would at least fail to adopt the belief that TRF is
Pyro-Glu-His-Pro-NH2. Moreover, we expect to find some kind of correlation
between the substances presented and the beliefs about structure that result,
so that the realist idea that beliefs covary to some extent with the state of
nature would be sustained. I find it implausible to think that, with respect to
this example, our judgment is entirely self-deception. The experiment we
envisage is too close in kind to naturally occurring situations in which people
who encounter different parts of nature form different beliefs.53
Let us consider some obvious objections. First, it may be suggested that
my formulation of the issue, with its pronouncements about what would have
happened had the world been different, is suspect. I am sympathetic to the
complaint that consideration of "ideal experiments" is somewhat artificial,
but it is important to see that such consideration crystallizes what is at stake
in the sociological critique. It is not enough simply to observe that there are
social processes that occur when consensus practice is shifted: the thesis is
that the character of the shifts is determined by these processes and not by
inputs from nature. That thesis immediately makes counterfactual claims. It
implies that differences in interactions with nature would not systematically
be reflected in the modifications of consensus practice. Talk of "ideal exper-
iments" and their consequences presents the counterfactual claims precisely.
Of course, we predict the outcomes of these "ideal experiments" by relying
on our judgments about what actual figures do in a variety of actual situations.
My formulations should appear less contrived if we recognize their kinship
with everyday knowledge about how people in different social contexts and
in different relations to nature behave, and how this knowledge serves us as
the basis for making counterfactual predictions: judgments of the forms "If
53. It seems to me that Latour and Woolgar, like many contemporary sociologists of
science, overestimate the power of the point that observation is theory-laden. Building on
the work of Hanson, Kuhn, Feyerabend, and Hesse, they advance what Michael Bishop
aptly calls "the expectation argument" because perception is theory-laden we see just
what our theoretical commitments would lead us to expect (Bishop forthcoming). But the
expectation argument is a gross hyperextension of what philosophers and psychologists are
able to show. Recall Kuhn's eminently sensible conclusion that anomalies emerge in the
course of normal science: Roentgen did not expect to discover X-rays (Kuhn 1962/70
57-59).
168 The Advancement of Science
S had not seen X then S would not have believed that p" and "If 5 had not
been in a society of type Y then S would not have believed that q" are hardly
alien. The suggestions about the outcomes of "ideal experiments" that I have
offered simply introduce such forms of judgment into reflection about episodes
of scientific change.
In some instances there are theoretical reasons for worrying that encoun-
ters with nature have no power to constrain consensus practice. Sociologists
of science have been much influenced by the thesis of the theory-ladenness
of observation and by the thesis of the underdetermination of theory by
evidence.54 They have also been much concerned with unsettled controversies,
in which, because we do not know how to describe the relations between the
participants and external nature, we are not able to use claims about the
differences in inputs from reality to explain the variation in belief. But to
recognize the methodological usefulness of ignoring scientists' encounters with
nature in certain types of analysis, should not lead us to the far more ambitious
claim that the world never has an impact on our beliefs. The inferential gap
is bridged by taking it for granted that there are alternative ways of responding
to any possible sequence of encounters with nature, so that the choice between
rival modifications of consensus practice must be determined by social forces.
A sociologist of science might contend that the (functional) cognitive
propensities present in mid-nineteenth century naturalists yield no resolution
of the question whether organisms are related by descent with modification,
given any possible sequence of encounters with nature that those naturalists
might have had, so that the actual resolution of that issue had to be effected
by social causes. That contention cannot be supported simply by pointing out
that there is an interesting network of social relations among the early de-
fenders of minimal Darwinism. The issue revolves around the possibility of
identifying a reliable process of reasoning that could generate the switch to
minimal Darwinism, given the stimuli that the naturalists actually received.
In Chapter 2 I tried to show that this is indeed possible, and that the process
is encapsulated in the long argument of the Origin (improved through the
social interactions of those who debated the Origin). I now suggest that this
serves as a paradigm for responding to sociological critiques. The first task
must be to formulate those critiques as I have done, in terms of "ideal ex-
periments" and the outcomes that both realists and their critics expect. Next
one can try to bring to bear empirical information about people's actual
behavior in different contexts on study of the conditions of decision. If it is
suggested that general considerationsfor example that of underdetermi-
nation of theory by evidencereveal that the scientific decision under scrutiny
could not have been determined through response to nature, then one must
consider whether or not there are available, reliable processes that would
54. Both the theory-ladenness thesis and the underdetermination thesis will be discussed
in Chapter 7. There is an obvious sense in which my reply to the arguments of Latour,
Woolgar, and especially Barnes and Bloor, is incomplete until something has been done
to defuse their claims about the radical underdetermination of our beliefs by all our en-
counters with nature.
Realism and Scientific Progress 169
deliver a verdict. Here there can be no substitute for looking at the historical
details of the circumstances under which decisions were made.55
8. Metaphysical Options
Realism is sometimes a modest doctrine. Much of the debate about scientific
realism in recent years has concerned a minimal version, the thesis that the
terms of successful scientific theories refer to entities that exist independently
of us. This version, entity realism, hardly seems worth the fuss. Unless we
are quite wrong about the simple (Lockean) idea that we interact with some-
thing independent of ourselves, then to the extent that references of our terms
are fixed through causal interactions, it is hard to see how we could not be
referentially successful. Of course, this is quite compatible with the thesis that
we know nothing at all about the entities to which we refer and, therefore,
seems of little comfort to a realist.56 Entity realism deserves Michael Devitt's
gibe that it is antirealism with a fig leaf (Devitt 1984 15).
I have spent most of this chapter replying to objections to a much stronger
version of realism, one that insists that many of the claims we make about
nature are true. Traditional critiques of realism are typically directed at this
stronger version (or particular articulations of it). However, the point of the
enterprise is to guard against challenges to the view of progress that I advanced
in the previous chapter. That view appears to be committed to an even bolder
version of realism, so that even if I have been successful in turning back the
complaints that have figured in recent discussions, the immodesty of my
position requires additional defense.
On my account the primary modes of scientific progress are conceptual
and explanatory progress. Both involve notions that may make modest realists
queasy. Is it reasonable to believe that there are natural kinds that exist
independently of human cognition? Or that there are objective dependencies
among the phenomena? Affirmative answers to these questions generate
55. The most challenging version of a sociological critique of the type of realism that
I favor is, in fact, extraordinarily rich in historical detail. In their (1985) Steven Shapin and
Simon Schaffer argue that the debate between Boyle and Hobbes over the "experimental
form of life" was only settled through "the acceptance of certain social and discursive
conventions, and that it depended upon the production and protection of a special form
of social organization" (1985 22). Because they are so convinced of the power of under-
determination arguments, Shapin and Schaffer develop their detailed history in a very
particular way, not focusing on the gritty details of the encounters with nature or the
complexities of the reasoning about a large mass of observations and experiments, but
making very clear the filiations to the social issues that concerned the protagonists. As I
shall suggest in Chapter 7, cases of putative underdetermination need to be handled ex-
tremely carefully, and, when that is done, Shapin and Schaffer's thesis becomes implausible.
But, precisely because of the richness of their studyand because of its many insightsa
complete reply would require a much longer discussion than this book can contain.
56. As noted in the last section, minimal realism is favored by several sociologists of
science. See, for example, (Barnes 1992).
170 The Advancement of Science
57. The general difficulty is exemplified in the dilemma posed by Paul Benacerraf (1973)
for theories of mathematical truth and mathematical knowledge. Benacerraf argued co-
gently that those accounts of mathematics that most obviously succeed in handling the
semantics of mathematical statements, providing an unproblematic account of mathematical
truth, strongly realist (Platonist) proposals that take mathematics to describe a mind-
independent realm of abstract objects, bring epistemological problems in their train. How
do we come into epistemic contact with the alleged objects, since causal contact is apparently
denied us? By contrast, those approaches that seem to dissolve the epistemological mysteries
(for example proposals that mathematics consists in recognizing the consequences of stip-
ulations that fix the meanings of mathematical terms) have trouble in developing an account
of mathematical truth. How does stipulation yield any type of truth? In general, realists
find it easier to define and explain semantical and metaphysical notions but then face
problems in understanding how we know what we seem to know.
58. See (Hempel 1965, Mackie 1973) for expositions of empiricist programs. Incisive
critiques can be found in (Earman 1986 chapter V, Lewis 1973a, and Salmon 1984, 1989).
59. The problem was explicitly posed, and linked to Benacerraf's dilemma, in (Mon-
dadori and Morton 1976).
60. David Armstrong responded to criticisms of Wesley Salmon's attempt to understand
our knowledge of causal claims (at a symposium on (Salmon 1984) held in 1985) by declaring
forthrightly that the epistemological problems were unworrying because "you can just see
causation."
Realism and Scientific Progress 171
63. For some of the complications involved in stating this principle of justified change,
see section 7 of my (1989). At this stage, since I only aim to sketch an alternative to strong
realism, an imprecise version of the principle will be enough.
64. See (F. Darwin 1888 II 13, 29, 78-79, 110, 121-122, 210-211, 240, 285, 327, 355,
362; III 25, 74; F. Darwin 1903 I 139-140, 150, 156, 184).
65. See (Kant 1781/1968, Sellars 1967, Putnam 1981).
66. If this is so, then it is easy to understand the role that appeals to potential and
actual unification play in scientific change. Darwin's stress on the unity that his modifications
of practice would confer is directly linked to the attainment of a fundamental aim. By
contrast, proponents of strong realism cannot take unification as constitutive of causal
structure, so that for them there are serious epistemological questions about whether the
search for unity will be a reliable guide to the causal structure of the world. This, of course,
is reminiscent of Benacerraf's dilemma in the case of mathematics. I have argued the point
in some detail in connection with Wesley Salmon's "ontic conception" of explanation. See
my (1989 section 8, especially 496-497).
Realism and Scientific Progress 173
9. Possible Losses
I began by considering a general worry about the coherence of the goal of
attaining truth, a worry that is prominent in Kuhn. I shall conclude with a
second type of Kuhnian concern, one that challenges ideas of accumulation
in science. After offering his traditional philosophical question about truth,
Kuhn advances his evaluation "as a historian" of the view that science makes
progress by achieving truth:
I do not doubt, for example, that Newton's mechanics improves on Aristotle's
and that Einstein's improves on Newton's as instruments for puzzle-solving.
But I can see in their succession no coherent direction of ontological devel-
opment. On the contrary, in some important respects, though by no means
in all, Einstein's general theory of relativity is closer to Aristotle's than either
of them is to Newton's. (1962/1970 206-207)
In other places, Kuhn is more explicit about the exact nature of the similarities
and differences. He claims that there are important affinities between Aris-
totelian explanations and explanations in nineteenth century physics:
As [physics] became increasingly mathematical, explanation came increas-
ingly to depend upon the exhibition of suitable forms and the derivation of
their consequences. In structure, though not in substance, explanation was
again that of Aristotelian physics. Asked to explain a particular natural
phenomenon, the physicist would write down an appropriate differential
equation and deduce from it, perhaps conjoined with specified boundary
conditions, the phenomenon in question. (1977 26)
The contrast is with the causal-mechanical style of explanation characteristic
of Cartesian and Newtonian physics, where, it is supposed, there are "active
agents" or "isolable causes."
67. Or, given my gloss on how the limiting of practices is to work, in all practices beyond
a particular point in a properly developed sequence. Here the notion of proper development
of a sequence of practices is partially understood in terms of the attempt to conform to
the principle of unification.
174 The Advancement of Science
NATURAL MOTION
Question: Why does body X move naturally to place P?
Answer:
(1) X is composed of the elements earth, air, fire, and water in the
proportions p : q : r : s.
(2) The natural place of a body composed of the elements in proportions
p : q : r : s is P(p, q, r, s).
(3) Bodies realize their natures most fully when they are in their natural
places.
(4) Bodies in natural motion strive to realize their natures most fully.
Hence (5) In natural motion, X moves to P.
GRAVITATION
Question: Why is X at (x1t*, x2t*, x 3t *) at time t*?
Answer:
(1) X is initially at (x10, x20, x30). The only other bodies in the universe
are Y1,. . . Yn with initial positions (y110, y120, y130), . . . .
(2) The mass of X is m; the masses of Y1, . . . , Yn are ml, . . . , mn.
(3) The only force acting is that of gravity.
(4) The force on a body of mass Ml from a body of mass M2 due to
gravity is attractive, directed along the line between the bodies, and is
of magnitude GM1M2/r2 where G is a constant and r is the distance
between the bodies.
(5) The equation of motion for a body of mass M subjected to a net
force Fi along the direction of the ith spatial coordinate is M d2xi/dt2 =
Fi
68. There are general suggestions about the natural places of "mixed bodies" in Ar-
istotle's Physics, but nothing as specific as the principle I ascribe. That, of course, is hardly
surprising given Aristotle's views about the distortions introduced by mathematization.
176 The Advancement of Science
(6) For the n + 1 body system consisting of X,Y1, . . . ,Yn the equations
of motion are. . . . [obtained by computing net gravitational forces de-
pendent on the relative positions of the bodies; there will be 3(n + 1)
simultaneous differential equations in 3(n + 1) coordinate variables and
time].
(7) The explicit motion of X is given by the equations xi = gi(t). [Ob-
tained by solving the system of equations (6) and using the initial con-
dition (1)].69
(8) gi(t*) = xit*.
Obviously this schema is very vague. There are correct versions of it, present
in Einsteinian physics and absent in Newtonian physics. Are any such versions
also present in Aristotle? That seems to me highly implausible. Aristotle
would not recognize the attributes of space that current physics takes to be
relevant to the gravitational field. Thus the most we can claim is that Aristotle
69. Of course, this is bluster when n > 1, since for three or more bodies the equations
are not generally analytically solvable. However, that point is irrelevant to the present
discussion.
Realism and Scientific Progress 177
1. Good Design
Before the early 1960s, almost all philosophers took the rationality of science
for granted. So prevalent was the notion that science is the epitome of human
rationality that explicit announcement of allegiance to it was largely confined
to popular lectures, introductory books, or preambles to more serious epis-
temological work.1 To account for this admirable feature of science it was
presupposed that scientists tacitly know methodological rules (jointly com-
prising "scientific method") which are used to appraise newly introduced
hypotheses and theories. Identifying these rules was an importantpossibly
the most importanttask for philosophy of science.2
In the next three chapters I shall try to see what, if anything, can be
salvaged from the thesis of the rationality of science. I shall begin by returning
the notion of rationality to its psychological roots. Science is not the sort of
thing that uses reasons, good or bad. Scientists engage in reasoning. The thesis
of the rationality of science ought to be interpreted as claiming that scientists,
for the most part or, perhaps, at the most crucial times, base their conclusions
on good reasoning. Or perhaps it intends to go deeper and trace the perfor-
mance of scientists to something in the institution itself that encourages good
reasoning.
The discussions of the last three chapters provide a way of thinking about
science that enables us to develop this approach. The science of a time is
1. See, for example, Reichenbach (1938), Hempel (1966), and Popper (1959). In all
these works, the hard problem is to identify the method that is taken to be at the core of
scientific rationality.
2. Many of the consequences of the older traditions become visible in the reactions
to Kuhn and Feyerabend. See, for example, Lakatos (1969), Scheffler (1967), Laudan
(1978), and, for an especially explicit version, Newton-Smith (1981). An alternative to the
implicit psychological hypothesis that scientists know and apply certain special rules is
Merton's (1942/1973) suggestion that features of the community of scientists account for
the choices made among theories, and thus for the rationality and progressiveness of science.
This kind of sociological approach was not developed by the leading defenders of ration-
alism. However, some proponents of evolutionary epistemology can be viewed as defending
the rationality of science without assuming the rationality of scientists. (See, for example,
Hull 1988.)
178
Dissolving Rationality 179
3. This is, of course, quite at odds with many discussions of rationality, both celebratory
and critical. Frequently, rationality is taken to be constituted by a set of rules whose status
is independent of their tendency to promote any ends. Explanations by appeal to rationality
require showing only that a particular set of statements conforms to the rules. As I shall
suggest, this flawed apsychologistic conception of epistemology invites the critiques of David
Bloor, Barry Barnes, and others. Moreover, the divorcing of rationality from any tendency
to promote epistemic ends fosters relativism. So, for example, the suggestion by P. F.
Strawson (1952) and Rudolf Carnap (1952) that the Humean problem of induction can be
solved/dissolved by taking inductive logic to comprise a system of principles that is con-
stitutive of rationality is rightly criticized by pointing out that there might be rival systems
of principles, espoused by different communities, that were regarded by members of those
communities as regulating the adjustment of belief. (Interestingly, this point was made
powerfully by a number of philosophers who were concerned with the "foundations of
induction." See, for example, (Feigl 1950), (Salmon 1968 33ff.) and (Skyrms 1986 Chap-
ter 2).)
4. Ronald Giere (1988) deliberately contrasts the kinds of evaluative projects of which
he approves (thinking about the adjustment of means to ends) with what he views as the
"traditional" philosophical conception of rationality (belied by the articles by Feigl and
Salmon, cited in the previous footnote.) Richard Foley's (1988a) not only explicates the
means-end approach but shows how long and rich is its philosophical history. Giere's
contrast thus seems to me to illustrate the way in which philosophical discussions can be
influenced by recent fashions, fashions that become canonized as "the tradition."
However, as John Worrall forcefully pointed out to me, a conception of rationality that
is solely concerned with the adjustment of means to ends and agnostic about the ends
would not capture many important themes that rationalistic philosophers of science have
180 The Advancement of Science
2. Illustration
At the 1860 meeting of the British Association for the Advancement of Sci-
ence, Robert Fitzroy, onetime Captain of HMS Beagle and former friend and
mess-mate of Charles Darwin, stalked the hall, holding a Bible above his
head and intoning the words "The Book! The Book!" Meanwhile, up on the
platform, the Bishop of Oxford, Samuel ("Soapy Sam") Wilberforce, engaged
5. Here I use the traditional language. As we shall see later in this chapter, the epithet
is only appropriate for at most one of the participants (Fitzroy).
182 The Advancement of Science
3. Preliminary Clarifications
Because debates about scientific rationality have often taken for granted a
very different perspective from the one I adopt, there is real danger that
firmly entrenched ideas about rationality will be carried over in thinking about
my conception of rationality as good cognitive design. Accordingly there are
some misunderstandings about past attempts to discern rationality in science
that it is important to correct and some criticisms of any such enterprise that
ought to be answered.
Defenders of the rationality of science are sometimes lumbered with the
doctrine that only those who hold true beliefs can be rational. Thus Bruno
Latour imagines attempting to divide figures from the past and the present
into two classes, one labeled with "derogatory adjectives" (of which 'irra-
tional' is a paradigm) and the other with "laudatory ones" ('rational,' for
example). The attempt, Latour claims, is doomed. Descartes initially appears
to be on the right sideuntil we recognize his commitment to bizarre vortices.
Newton only lasts among the praiseworthy while we keep his alchemical
researches under wraps. So it goes. Latour concludes wryly: "The only way
to stop adjectives jumping randomly from one side of the divide to the other
would be to believe that only this year's scientists are right, sceptical, logical,
etc." (1987 191).
Well, of course one can make the enterprise of appraisals of rationality
look foolish and unworkable by choosing the wrong units of analysis. Whoever
thought that a person would be rational or irrational throughout her entire
career? Indeed, the tradition aimed at specifying the rationality or irrationality
of particular beliefs or decisions, held or made by particular people in par-
ticular contexts. The appropriate candidates for division are not therefore
people (Newton, Descartes) but the beliefs they held or the things they did
Dissolving Rationality 183
12. Simple examples are readily generated by thinking about situations in which stu-
dents (from very young children to apprentice scientists) are tested. Of course there are
many instances in which students arrive at the right answer but are properly corrected for
having engaged in reasoning that is unreliable (likely to generate false answers, and only
successful on the present occasion as a matter of luck). While these examples are distant
from the more complex situations of scientific debate, so that there is a genuine issue about
whether or not the distinctions of good from bad cognitive design carry over to those
contexts, they do demonstrate the possibility of honoring the idea that beliefs are part of
the natural order and simultaneously pursuing normative questions.
186 The Advancement of Science
ceiving badly, inferring hastily, failing to act to obtain inputs from nature that
would guide them to improved cognitive states. Nobody who accepts the
symmetry thesis, who believes (rightly) that changes in cognitive state have
causal explanations, should go on to infer that there is no basis for distin-
guishing the types of processes that generate and sustain beliefs, actions, and
decisions. Some types of processes are conducive to cognitive progress; others
are not.
From this perspective Latour's dismissal of talk about rationality disguises
the hard problems by suggesting that we can and do make judgments about
rationality in any way we like. He thinks that we can vary our perspective
on Galileo, seeing him alternatively as "courageously rejecting the shackles
of authority, arriv[ing] at his mathematical law of falling bodies on purely
scientific grounds," or as "a fanatic fellow traveller of protestants [who] de-
duces from abstract mathematics an utterly unscientific law of falling bodies"
(1987 191, 192). However, if our aim is to do more than make up stories
about Galileo and we proceed with the approach to rationality I have been
recommending, then we face two difficult projects. First we must identify the
processes that underlie Galileo's beliefs and decisions. Second we have to
assess the ability of such processes to generate and sustain cognitively valuable
states. Both endeavors are error-prone. But there is a fact of the matter:
either Galileo's reasoning exemplified a strategy likely to promote cognitive
goals or it did not.
Why should we engage in such difficult endeavors? Since Bacon and Des-
cartes, explicit methodological reflection has been designed to improve human
cognitive performance, and we look to the apparent successes of the past for
guidance.13 So the (fallible) enterprise of evaluating the rationality of our
predecessors does not simply express our penchant for grading people, award-
ing gold stars here and black marks there. Methodologists hope to analyze
successful scientific reasoning in order to integrate those cognitive strategies
that promote progress within the science of the present. Psychological research
not only demonstrates that subjects have difficulty assimilating certain types
of cognitive strategies (appealing to base rates, for example) but also shows
that successful strategies can sometimes be absorbed into their cognitive lives
(Holland et al. 1986 260-271). The delineation of formal rules, principles,
and even informal canons of reasoning, when supplemented by an appropriate
educational regime, can thus make people more likely to activate propensities
and undergo processes that promote cognitive progress.14
13. Here, and elsewhere in Chapters 6, 7, and 8, I emphasize the meliorative project
of epistemology. However, as both Peter Godfrey-Smith and Kim Sterelny have pointed
out to me, epistemologists are rightly also interested in understanding the extent to which
human cognitive performance is successful. Given my account of cognitive goals in Chapter
4, it would be ironic if I were to slight the significance of understanding, whether or not
it leads to improved performance. Thus my claims about the meliorative project should be
seen against the background of a broader epistemic enterprise.
14. Consider one of the canonical problems for which Bayesian analysis is useful, the
computation of risks in cases where a medical diagnostic test has a known rate of false
Dissolving Rationality 187
positive results. If you are given a test for a rare disease that gives a small rate of false
positives, and if you test positive then your risk is correctly assessed by employing infor-
mation about the base rate and using Bayes's theorem. If you ignore the base rate and
follow some non-Bayesian process of reasoning you will almost certainly arrive at erroneous
conclusions about your risk. Granting that people do not naturally perform the Bayesian
computation, the significant points are that that computation improves their decision making
and that they can be taught to do it.
15. It is important to distinguish here between the "nuts-and-bolts" methodological
advice about experimental design and statistical analyses of data that is genuinely useful
for scientists, and the more grandiose methodological pronouncements ("Always try to
falsify your pet hypothesis!," "Prefer simpler theories!") that are sometimes adapted from
philosophy to provide exhortations to the scientific apprentice. These slogans, often at odds
with scientific practice, should no more be taken as philosophy's main contribution than
should the historical anecdotes that leaven scientific texts be seen as history's gift to science.
In both cases, simple stories and simple slogans may provide rough and ready pictures,
valuable to the beginner but unneeded (and discarded) by the veteran. However, even the
veteran may sometimes be well served by historians and philosophers who can provide
nuanced accounts that are relevant to current debates. Recent discussions in evolutionary
theory have drawn, sometimes substantially, on historians' reconstructions (see, for ex-
ample, Gould 1977, Eldredge 1985) and philosophers' attempts to articulate methodological
ideas that are directly relevant to current debates (Sober 1984, 1988, Brandon and Burian
1984, Kitcher 1985b, Dupre 1988).
188 The Advancement of Science
chapter will lay any such concerns to rest. In the meantime, it is possible to
explain in general how the thinking and decision making of the past can be
evaluated at least partly on their own terms.
To evaluate a scientist's efforts to achieve cognitive goals, we need a view
of the scientist's situation, a view of the scientist's capacities, and a view of
what would have promoted attainment of the goal. These views should draw
on the best available information. Acquiescing in current science, we use
contemporary knowledge, including historical knowledge, to make our as-
sessment. Does this mean that we necessarily criticize the scientist for failing
to see what we see or for failing to undergo the processes that we believe to
promote cognitive progress? Of course not. We use our knowledge of human
cognitive systems and of the historical situation to identify the kinds of pro-
pensities that would have been present in our subject and thus to delimit the
kinds of psychological processes that might have been available to modify the
cognitive state. We employ our current understanding of the world to consider
the relative abilities of such processes to promote cognitive progress. Cog-
nitively well designed scientists use processes that are relatively good at pro-
moting cognitive progress, within the comparison class of those that are
available to them. They do not have to believe what we believe or even to
reason as we would reason. They are seen as optimizing (or, at least, doing
relatively well) subject to constraints (often constraints from which we are
liberated). They do the best they can.
This general explanation shows that the more blatant forms of historical
chauvinism can be avoided. We do not have to assume that only those who
share our beliefs or emulate our strategies of good reasoning count as rational.
But it is perfectly reasonable to wonder whether subtler versions have been
circumvented. As we shall see in the next two sections, the notion of good
cognitive design needs to be carefully disambiguated before such issues can
be resolved.
(ES) The shift from one individual practice to another was rational if
and only if the process through which the shift was made has a success
ratio at least as high as that of any other process used by human beings
(past, present, and future) across the set of epistemic contexts that in-
cludes all possible combinations of possible initial practices (for human
beings) and possible stimuli (given the world as it is and the characteristics
of the human recipient).
16. Ideally, we would like to delimit the strategies that are psychologically possible for
members of our species. Those that are actually used comprise a proper subset of these.
Moreover, it may well be that there are some strategies that would be excellent and would
be employable by us, but which are never actually going to be used. For present purposes
my concern is to provide a relatively sharp standard of rationality, so I shall resist the
temptation to take the notion of "psychological possibility" or "availability" as well under-
stood, even though there is a danger that the class of strategies on which I focus may
seriously underrepresent the target.
190 The Advancement of Science
that a strategy be optimal for doing a particular kind of job, rather than
optimal simpliciter.) The set A could be indexed to the cognitive resources
available at the time of the subject whose rationality we are considering, so
that we exclude processes that would have been simply inaccessible to people
at that timeprocesses that only become available through the explicit artic-
ulation and inculcation of methodological principles, for example. We might
also decide that success ratios over the entire set C are not of interest to us,
limiting our attention to the kinds of practices that are (in some sense) avail-
able to the contemporaries of our subject and to the kinds of stimuli that they
might have received. Unlike the earlier proposals for modifying (ES), this
one does not necessarily weaken the standardthe set of shifts it would admit
as rational does not necessarily include those admitted by (ES). Similarly, if
we amend the measurement criterion by taking into account the magnitude
as well as the direction of the change in practice, we are likely to obtain a
different standard but one that may not be any weaker.17
Judgments about the history of the sciences often resolve the vagueness
in context by implicitly comparing two rival processes. Simple claims about
rationality can avoid fine-grained resolutions by focusing on the differences
between the processes undergone by those whom we retrospectively praise
and the processes of their opponents. Galileo's modification of his practice
counts as rational because looking at the heavens rather than not looking is,
it seems, likely to have a high success ratio in modifying astronomical practice.
The details of which processes are available, the epistemic contexts we con-
sider, the ways cognitive improvement is measured seem insignificant beside
the gross differences we expect to find. However we do the accounting we
think that we shall get the same result. Of course, as the history becomes
more refined and it is understood why the telescope might have seemed an
unreliable instrument, confidence that we can ignore the vagueness begins to
waver.
Which standard is the right one? There is no answer. A methodological
ideal would be to describe optimal improving processes for as broad a range
of contexts and as large as class of alternatives as possible. (ES) provides a
valuable target for the methodologist: pointing to a goaloptimal cognitive
designat which we aim. Where explicit methodology is done, one can thus
expect that there will be attempts to meet the exacting demands of (ES).18
17. However, there is a reasonable worry that (ES) as formulated is too demanding
because of the presence of relatively trivial processes of modification with success ratios
close to 1. Consider the possibility of always effecting a small improvement in practice by
carrying out some deduction from "safe" premises. This would, presumably, yield minuscule
improvements in practice across almost all contexts (the exceptions being those in which
rigorous house cleaning was in order). Risky strategies for making major improvements in
practice would thus be debarred because they would have lower success ratios. This problem
can obviously be mitigated by taking into account the magnitude as well as the direction
of the shift, so that we allow for processes that occasionally yield big dividends.
18. Or of some kindred principle of maximization that takes into account not only the
fact that processes generate positive modifications but also the extent to which they promote
cognitive progress.
Dissolving Rationality 191
suited for the generation of cognitive progress. Simply lumping these various
kinds of mistakes together with one another and with the extreme lapses of
those who serve us as paradigms of irrationality is unhelpful. It is possible to
introduce a finer-grained taxonomy of cases by recognizing that propensities
for reaching decisions by undergoing certain kinds of processes, some of which
would be progress-promoting, can also lead to performances that go astray
in various ways. So, for example, my imaginary Bayesian bungler activates
a good higher-order propensity, trying to resolve a problem by applying Bayes-
ian ideas in a case to which such ideas are appropriate. The goodness of that
propensity consists in the fact that the processes it can be expected to produce
include some that would promote cognitive progress, even though the actual
process generated is not among them. Similarly, in instances of complex
scientific evaluation, proponents of rival positions may attempt to assemble
evidence, assess arguments, and apply methodological principles in an overall
fashion that would promote cognitive progress, despite the fact that their
actual performances are flawed. We may applaud the basic propensities while
recognizing the foibles of the resultant processes.
Turn now to my second scientist. Evaluating her epistemic performance
depends on the details of the story. The appeal to tradition may sometimes
be born of inflexible, dogmatic thinking (of this, more later). It may also arise
from modesty (or from the fact that not everything can be checked at once).
Once we have broken free of the idea that methodological principles can be
known a prioriso that one can know, just by thinking hard enough, how
to satisfy (ES)then methodological misinformation, sanctioned by tradition,
can be treated in the same fashion as substantive misinformation, encapsulated
in hallowed lore. Scientists may tentatively explore a line of reasoning that
would, as things stand, promote cognitive progress, drawing back from it
because they perceive it as at odds with approved canons. Their rejection
may be deliberately based on a sense of the improbability that they will have
made a methodological breakthrough that their wise predecessors and con-
temporaries have missed. Something must be wrong with the apparently ap-
pealing argument. So they reason, and, in following traditional ideas, they
undergo a process that is far less well designed for promoting cognitive
progress.
Let us take stock. I began by approaching scientific rationality from the
perspective of the external standard: rational decisions are those that issue
from processes that have a high expectation of promoting cognitive progress.
But there is no single external standard. Rather, there are many different
ways in which we may make the general idea more precise. However, sat-
isfaction of the external standard, even in its most exacting form, does not
guarantee that all kinds of cognitive lapses have been avoided, for there may
be people who meet that standard fortuitously by following bizarre higher-
order strategies. On the other hand, sometimes those who are moved by
higher-order propensities that have a high expectation of generating processes
that meet the external standard should count as rational, despite the fact that
the actual processes that underlie their decisions do not meet the standard.
194 The Advancement of Science
20. For an account of the origins of Galileo's troubles, see (De Santillana 1955); for
radically different suggestions, see (Redondi 1987). (Finocchiaro 1989) provides an ex-
tremely useful set of primary sources for understanding the Galileo affair.
Dissolving Rationality 195
twenty years, the major exponents of creation science have been declaring
that the second law of thermodynamics is incompatible with the evolution of
life. Creationists have been in the presence of people who have given lengthy
critiques of their objection and there is substantial evidence that their eyes
have wandered over some of the pages on which such critiques have been
printed. How has their thinking adapted to these critiques? Apparently not
at all, for they make no replies to them and continue to present their ideas
in exactly the same ways.
There are limits to proper tolerance. In some cases, epistemic perfor-
mance is so inflexible that we either view the cognitive systems in question
as poorly designed for the promotion of cognitive goals or suppose that the
goals that are being activated are not cognitive at all. Where the latter
supposition is correct and the subjects in question profess cognitive goals,
some form of deception or self-deception is occurring. The category of
pseudoscientists is a psychological category. The derivative category of
pseudosciences is derivatively psychological, not logical as philosophers have
traditionally supposed. Pseudoscientists are those whose psychological lives
are configured in a particular way. Pseudoscience is just what these people
do.
6. Closing Debate
Dissolving rationality, we have refocused the issues to which that notion was
directed in terms of goodness of cognitive design. Let us now turn back to
the question of the "rationality of science." In holding science to be the
supreme embodiment of human reason, rationalists21 are advancing claims
about the ways in which scientific debatesincluding those debates that mark
times of large transitionhave typically, if not invariably, been closed. Their
antirationalist opponents offer rival ideas about the closure of those debates.
I shall start by showing how the fundamental themes of rationalism and
antirationalism can be expressed in my preferred idiom.
The rationalist model embodies the following assumptions:
(R1) The community decision is reached when all individuals within the
community have independently made the same modification of their
practice.
(R2) Each member of the community is moved solely by the epistemic
goal of modifying practice as progressively as possible.
(R3) All members of the community are in the same epistemic context:
each begins from the same practice and each receives the same stimuli.
21. Here, and in what follows, I use this term to designate those who believe that
scientific debates are closed through the formulation and appreciation of decisive reasons.
Dissolving Rationality 197
(R4) While there is debate within the community,22 those who champion
the modification that ultimately triumphs do so by undergoing processes
that are well designed for promoting cognitive progress (processes that
satisfy (ES) or some such standard); their opponents undergo processes
that are less well designed for promoting cognitive progress.
(R5) Debate closes when those who employed the inferior processes
rightly modify their cognitive activity so as to undergo the progress-
promoting processes; in some instances, a small minority may fail to
make this shift and members of such minorities are excluded from the
pertinent communities.
22. The extreme version of rationalism supposes that, from the first public introduction
of the ultimately successful view, those who adopt it do so because they undergo cognitively
superior processes. So, for example, Copernicanism would have been epistemically pref-
erable as of 1543, Darwinism as of 1859, special relativity as of 1905. There are often
sustained efforts to show that the enlightened were thinking clearly from the beginning
see, for example, the treatments of Copernicanism by Glymour (1980) and by Lakatos and
Zahar (1976). As I shall suggest, this is a rationalist theme that ought to be dropped. Just
as rationalists should be happy with the possibility that those who break with tradition to
develop completely new viewsCopernicus in 1509, Darwin in 1838take large epistemic
risks, so too, they should allow that early adherence by others need not be on the basis of
undergoing cognitively superior processes. Once the issue is posed, many erstwhile ration-
alists may prefer to remain more flexible about the time at which cognitively superior
processes become available within the community, maintaining only that this occurs before
closure of the debate.
198 The Advancement of Science
23. In denying (R4), antirationalists also raise to prominence features of the community
of scientists that are explicitly ignored in (R1)-(R3). They should thus be viewed as denying
the validity of rationalist idealizations.
24. This thesis encapsulates the popular idea of underdetermination of theory by evi-
dence (see Chapter 7). A more extreme version of (AR4) would claim that the processes
undergone by the ultimate victors are less well designed for promoting cognitive progress
than those employed by the ultimate losers. Thus Feyerabend takes up Galileo's praise of
Copernicus for denying reason. However, the ultimate point of Feyerabend's argument
seems to be that no coherent general account of progress-promoting reasoning can be given,
so I shall suppose that he, like other antirationalists, would be content with (AR4).
Dissolving Rationality 199
25. One important way in which a debate can evolve is if the points in dispute are
significantly modified. Later in this chapter I shall discuss an example in which this happens
(the course of "the great Devonian controversy"). Here I am concerned with changing
argumentation directed at a pair of proposals for modifying practice that remain constant
throughout the period of debate.
26. This is because the victors have shaped the culture in which science is done, setting
the standards for which kinds of considerations will be taken as decisive.
200 The Advancement of Science
By supposing that superior progress-promoting processes only become avail-
able late in the course of the debate, one raises obvious questions about the
early champions of proposals for modifying practice. Their reasoning (more
generally, their cognitive activity) cannot be seen as superior to that of their
opponents. Yet out of their efforts, indeed perhaps out of interactions with
their critics, there emerges a line of argument that enables all members of
the community to undergo superior progress-promoting processes.
As I have suggested, and as (AR3) explicitly maintains, there is cognitive
variation within scientific communities. Scientists differ with respect to their
individual practices, with respect to the information they exchange with other
members of the community, in the stimuli they receive from nature, and in
their propensities for modifying their cognitive states. I shall suppose that it
is possible that, at early stages of scientific debate, the processes that underlie
the practices of differently situated members of the community should satisfy
some standard of good cognitive design (not necessarily the same in all cases)
even though those practices differ with respect to the issues involved in the
debate. In virtue of the adoption of those practices and the conversations
with peers and the encounters with nature that they generate, the participants
receive new stimuli. They modify their practices and the underlying processes
in ways that meet some standard of good cognitive design (again, not nec-
essarily the same in all cases). After a number of rounds of debate, matters
crystallize. Certain kinds of cognitive homogeneity have been achieved, cru-
cial issues and arguments formulated, and the product is a form of reasoning
that encapsulates a process that is markedly superior (judged by (ES) and
other standards of good cognitive design) both to the processes that sustain
the rival proposal(s) for modification of practice and to those that originally
underlay the successful change. Internalization of this form of argument by
most members of the community builds sufficient power among those who
champion its conclusion to dismiss dissenters from further discussion.
Traditional forms of rationalism about scientific change implicitly condemn
those who behave in the fashion that I have sketched. Rational scientists with
partial evidence are supposed first to assemble all relevant evidence and then
to believe exactly what is warranted by the total evidence; if there is a standoff,
then the proper scientific stance is agnosticism. Two important questions
emerge from the contrast between this picture and mine. First, is good cog-
nitive design always reflected in action to gather more evidence and agnos-
ticism when the evidence is balanced, or are risky cognitive processes
sometimes progress-promoting? Second, is it a good thing from the point of
view of the community that scientists are sometimes prepared to jump to
conclusions, pursuing lines of inquiry that lead to clarification of the issues
under debate? I shall examine both questions in the next two chapters and
(as may be expected) suggest that the answers to them favor my views rather
than those of the tradition.
I have been attempting to motivate a model for the closure of major
scientific debates that embodies some of the ideas of rationalism and some
of antirationalism. This compromise model comprises the following claims:
Dissolving Rationality 201
27. It is quite possible that erstwhile rationalists may regard this as the crucial point
and thus view the compromise model as entirely congenial. The model may be viewed as
making some rather obvious amendments to classical rationalism once the apsychologistic
approach to rationalism has been abandoned. So, once convinced of the need to do epis-
tcmology in psychologistic terms, a rationalist might regard (R1)-(R5) as obviously flawed
and take the compromise model to be the straightforward way of amending them. If so,
then, as I have argued in earlier sections, approaching epistemological questions psycho-
logistically is really the crucial step.
202 The Advancement of Science
factors and arguments leading in a contrary direction, the acquisition of power
should be more affected by the arguments. (Social factors may retard a de-
cision but not reverse it.28)
It may be too optimistic to think that every debate, even every major
debate, in the history of science has been resolved in accordance with the
compromise model. There may be occasions on which modifications of prac-
tice are made prematurely, so that it is only later, after further encounters
with nature, that cognitively superior processes are available to support the
modified practice.29 Would-be rationalists can tolerate lapses from the closure
of debate by dissemination of superior reasoning, if it can be shown that the
lapses do not matter.
Recall the point of insisting on the rationality of science. Those who
believe, as I do, that we discover more and more about the world while
simultaneously learning how to investigate the world are vulnerable to skep-
tical challenges to the effect that cardinal tenets of contemporary science are
mistaken. Rejecting the possibility of any a priori foundation either for science
or for methodology, we can only answer skeptics by pointing out that our
current knowledge is the product of a self-correcting process. Our right to
this defense is undermined if there are transitions in the history of science
which could have been made differently, on the basis of cognitively equivalent
processes, and which would have yielded very different contemporary prac-
tices. The simplest response to that deep skeptical complaint is to argue that
all transitions in the history of science accord with the compromise model.
Failing that, we can block the complaint by arguing that the cognitively su-
perior processes would have emerged later, whether or not the premature
decision had been made, or by showing that an alternative modification of
practice would ultimately have converged on the actual historical sequence
of practices. In both instances, there are substantial tasks to discharge. For
example, one must defuse the idea, bruited in (AR5), that decisions to modify
practice contaminate the reception of further stimuli, so that, once the pre-
mature decision has been reached, the scales are automatically tilted so as to
ensure its success according to any cognitive criterion, no matter how de-
manding. Similarly, showing how an alternative decision at some key point
would still eventually have generated practices indistinguishable from our
own plainly requires ingenuity in spinning historical fiction. Hence it is far
simpler to ward off skepticism by contending that transitions in the history
of science fit the compromise model. If, that is, the charms of simplicity and
historical accuracy can be combined.
In the next two chapters I shall explore individual and community decision
making with an eye to seeing how much of this residual rationalism can be
28. Although the compromise model has not been explicitly formulated before, I believe
that it has affinities with one of the positions about scientific change offered in (Kuhn 1962/
1970) and developed in chapter 13 of (Kuhn 1977). Certain forms of Bayesianism may also
find the compromise model congenial (see, for example, Salmon 1990).
29. Here, as elsewhere, I count an unaltered practice as undergoing a "null" mod-
ification.
Dissolving Rationality 203
7. Scientists at War
Scientific debates are rarely resolved in an instant. Most, if not all, of the
examples that have figured in rationalist reconstructions of the history of
science concern episodes in which over a period of years, even generations,
the merits of a particular way of modifying consensus practice are vigorously
debated. During these extended debates, the participants undergo innumer-
able cognitive processes as they struggle to assess the credentials of rival
proposals. Neither the proposals nor the arguments offered in their favor are
stable. It would thus be highly surprising if there were available to the par-
ticipants some simple way to reach a rational decision. Those who believe in
some clean criterion for distinguishing (say) Copernicanism from Ptolemaic
astronomythe progressiveness of Copernicanism as a research program or
the greater bootstrap testability of Copernicanism, for examplecan only
defend the rationality of the early Copernicans if they can show why it was
cognitively preferable for midsixteenth century people to focus the debate on
this criterion, ignoring the mass of other factors that occur in discussions of
De Revolutionibus.
Thinking about scientific controversies is considerably advanced by noting
their resemblances to war (or to a complex "war game" like grandmaster
chess). In scientific debate, as in warfare, patterns of alliance and opposition
may change. Just as in scientific debates the proposals defended and the issues
at stake may be altered, so too in warfare the demands and goals of the parties
may shift with time. Most importantly, wars are rarely resolved through some
single, local incident. There are occasional exceptionsperhaps the Norman
conquest of England can be traced to the arrow that killed Haroldbut, in
almost all cases, individual skirmishes do not decide battles, individual battles
do not determine the results of campaigns, and individual campaigns are not
always crucial to the outcome of a war.
In major scientific debates, we can also think in terms of skirmishes,
battles, even campaigns. Nor is this idiom foreign to the way in which the
scientists involved are prepared to talk: Darwin, recall, describes his book
on orchids as a "flank movement" against the enemy. Just as the ultimate
victors rarely win every battle (and may even, as the Russians against Na-
poleon, lose all the early battles), so too, the fortunes of scientific debate
may be variable. We can expect, then, that scientists who end up on the
losing side may still "fight well" (produce good reasoning) and may even win
some victories.
I propose to extend the compromise model by drawing on the resemblances
between science and warfare, and, in particular, underscoring the idea that
scientific controversies are long and complex episodes, in which local points
204 The Advancement of Science
and items of evidence do not determine the final result. The participants in
the debate defend proposals for modifying consensus practice by offering a
list of characteristics that a practice should have and presenting reasons for
thinking that that modification satisfies the desiderata or is likely eventually
to satisfy those desiderata. These defenses are attacked by criticizing the
desiderataeither by suggesting that important requisites of cognitive prog-
ress have been omitted or by contending that the alleged desiderata are
inappropriateor by objecting to the claims about the likelihood of satisfying
the desiderata. So there arise skirmishes about particular issues. A skirmish
is resolved in favor of the attacker if the defender acknowledges both that
there is a legitimate desideratum of doing X and that all of the specified
avenues for doing X seem to be blocked. A skirmish is resolved in favor of
the defender if the attacker concedes either that an alleged desideratum is
inappropriate or that the defender has specified a way to meet it. There are
a number of intermediate cases, in which the defender proposes one or more
possible avenues along which a solution to the problem of meeting the de-
sideratum might be found. These proposals may raise subsidiary questions
about the probability with which exploring those avenues will prove successful.
The weapons available for carrying out these skirmishes are an agreed-
upon conception of the goals of science, possibly only at the most abstract
level (that presented in Chapter 4), possibly involving agreement on derivative
goals as well, a consensus on certain elementary forms of argument, and
agreement about some statements.
I shall postpone to the next chapter the task of considering how individual
skirmishes are decided and how the outcomes of those skirmishes affect the
outcome of the debate. Here I shall only sketch some features that appear
in many scientific controversies. At early stages of debate matters are typically
confused. Each side is able to resolve some skirmishes in its own favor.
Confusion dissipates as one party manages to reopen the issues ("terrain")
lost in previously unfavorable skirmishes, now engaging in inconclusive skir-
mishes (by thinking of some hitherto unexplored avenues for meeting the
desideratum) or even in skirmishes that are resolved in its own favor.30 If this
successful defense is accompanied by an increase in the number of unanswered
attacks, then, just as in war or in chess, there is a clear sense of a position
crumbling.
Assuming that the forms of argument have desirable qualities (reliability
in generating true conclusions from true premisessee the next chapter) and
that the participants accept their common premises on the basis of processes
that promote cognitive progress, then the arguments they offer reflect pro-
cesses of reasoning that cannot be dismissed as poorly designed for the pro-
motion of cognitive progress. At early stages of the debate, the processes
underlying the claims of each side are risky: each has to admit that certain
30. Here, and elsewhere in this section, there is kinship with the ideas of Larry Laudan
in his (1977). Skirmishes are fought around what Laudan calls "problems." Both notions
deserve further analysis and explanation. My account will be offered in the next chapter.
Dissolving Rationality 205
problems will have to be overcome in ways that they are not yet able to
specify. As the debate crystallizes, one side finds avenues which can be ex-
plored to find solutions, and there emerges a progress-promoting process
which, once made available to all, inclines virtually all members of the com-
munity to the pertinent modification of practice. This is analogous to the
major battle, or final push, that concludes many wars.
proposes to explain the apparent motions in terms of the true motions and
to provide an account of why the true motions are as they are. Decisions
about the appropriate level for astronomy will be based on assessment of the
past history of efforts to solve various kinds of astronomical problems and
the likely possibilities for future solutions. Those who are relatively pessimistic
will settle for level 1 and appraise astronomical schemes solely in terms of
predictive accuracy and mathematical tractability. Considerations about the
possibility of the moving earth then become quite irrelevant and Coperni-
canism can be defended along the lines of Osiander's famous preface. For
those who adopt level 2, there will be complicated choices among Ptolemaic
astronomy, Copernican astronomy, or some other system, based on evalua-
tion of the competing advantages and problems of both, but, since the link
between astronomy and Aristotelian cosmology has been broken, certain
kinds of constraints from physics will no longer be in force. Finally, astron-
omers at level 3 will have an even more complex decision: they may try to
rebuild the old Ptolemaic-Aristotelian universe, embed Ptolemaic astronomy
within a new cosmology, or offer Copernicanism or some new system with
cosmological and physical underpinnings.
In the second half of the sixteenth century many of these options are
played out. Copernicanism is developed mathematically by astronomers at
level 1.34 It is largely rejected by astronomers at level 2, who either seek to
defend versions of Ptolemy or else attempt to build new systems (Tycho Brahe
is the most famous example).35 Kepler is the most prominent figure at level
3, and he, of course, opts for Copernicanism. I claim that none of the initial
decisions made by these men was irrational. In each case there is an adequate
process that we can presume to have generated their modifications of indi-
vidual practiceand, I suggest, there could have been adequate processes
that led to some of the unexplored options as well.
The situation changed in the early seventeenth century. Kepler's labors
issued in the production of a more accurate scheme of planetary astronomy
than any hitherto available (although Kepler's mode of presentation initially
prevented his ideas from securing a wide audience). At the same time, Gali-
leo's telescopic researches and his incisive critiques of aspects of Aristotelian
cosmology, physics, and methodology turned back the most important chal-
lenges to Copernicanism. The debate crystallizes in Galileo's famous Dialogue
Concerning the Two Chief World Systems, where the principal arguments for
rejecting Copernicanism are systematically evaluated.
The target of the Dialogue includes several of the options I have distin-
guished: Galileo attacks attempts to combine Ptolemaic astronomy with Ar-
istotelian cosmology and negative evaluations of Copernicanism that are based
on either Aristotelian physics or cosmology or commonsense ideas about
36. The approach to questions about underdetermination adopted here will be artic-
ulated further in the next chapter.
37. One important vehicle for strengthening the astronomical arguments was Kepler's
Epitome of Copernican Astronomy. Robert Westman has suggested to me that this work
was more influential in securing the acceptance of Copernicanism than was Galileo's Dia-
logue. For my purposes here, it is not important to divide the laurels between Kepler and
Galileo. The significant point is the emergence of a strong line of pro-Copernican reasoning
in the 1620s and 1630s.
210 The Advancement of Science
the midseventeenth century are no longer just the university men and (to a
lesser extent) the church scholars. Court patronage of both individuals and
groups has created networks of investigators who self-consciously deride the
stuffiness of traditional universities, and there are men of independent means
("gentlemen") who dedicate their resources to "philosophy." Meanwhile, of
course, the university curriculum has been under a broad attack, and, in some
universities, there has been reform of the traditional curriculum. These
changes are partly dependent on the course of the Copernican debate, partly
independent, but they create a social milieu in which the arguments of Galileo
secure a far more sympathetic hearing than they might have expected a half-
century before. While the closure of the controversy is, I believe, to be
understood in terms of the acquisition of epistemic authority by newly emerg-
ing groups on the basis of an appreciation of superior reasoning, it is important
to be aware both of the extent to which the set of groups with epistemic
power does not remain constant through the period and of the variety of
factors that play a role in the waxing and waning fortunes of various constit-
uencies.
The presence of nonepistemic interests, as claimed by (C2), is plain in the
concerns about the religious implications of Copernicanism, voiced by some
Lutherans in the midsixteenth century and by some Catholics in the early
seventeenth century. Worries about credit within the community of scholars
also surface. Tycho's bitter polemic with Ursus shows how deeply the former
was concerned to enhance his reputation. Galileo's quest for patronage reveals
both his interest in securing a livelihood that would give him the leisure for
his scientific researches and his attachment to la dolce vita.
Cognitive variation, as demanded by (C3), is also easy to find. Copernicus
himself made few observations but was steeped both in the technicalia of
astronomical geometry and in the humanistic tradition. Tycho's observations
were unparalleled among his contemporaries. Kepler's knowledge of optics,
harmonics, and numerology contributed to making his astronomical writings
virtually unreadable. In the century between Copernicus and Galileo, we find
numerous men struggling with the problems of the solar system; applying
different bodies of background information to different questions that they
hail as significant, with very different cognitive temperaments (contrast the
enthusiasm of Rheticus with the conservatism of Osiander); and reaching very
different conclusions. The Copernican debate is not simply a matter of he-
liocentrism against geocentrismwith the Tychonic system thrown in for good
measure. Almost every participant has unique ideas about which problems
are important and how those problems should be solved.
Full defense of the applicability of (C4) and (C5) is impossible without
more careful elaboration of the reasoning of the various participants and the
kind of analysis that I shall undertake in the next chapter. However, the main
lines of the defense are easily indicated. Given the astronomical advantages
of Copernicus's scheme and its physical disadvantages, four different re-
sponses are readily comprehensible. Perhaps the physical difficulties of Co-
pernicanism could be solved. Perhaps the Ptolemaic system could be revised
Dissolving Rationality 211
38. Obviously there are worries that the success of the model is an artefact of the precis
that I make. I shall do what I can to forestall such concerns, by indicating how my story
relates to Rudwick's far more complex narrative.
212 The Advancement of Science
39. As Rudwick makes very vivid for the reader, this concession caused De La Beche
considerable painnot least because he worried that admission of an error, even an ex-
cusable error of the type he had committed, would be interpreted by his governmental
superiors as a sign of incompetence, with dire consequences for the future of the publicly
funded survey and for his own employment. De La Beche was thus constrained by factors
external to the controversy, but, with the help of his friends, particularly Greenough, he
was able to accept the correction without losing professional credibility.
40. Another aspect of Rudwick's story that is slighted in my treatment is the way in
which the geologists most intimately involved in the controversy were forced to rely on the
classifications offered by the fossil experts. The available "relevant evidence" was broadly
distributed across the group of people involved, some of whom had at their fingertips
detailed knowledge of the Devon strata but little background either in general paleontology
or in taxonomy, others who had never seen the rocks in question but were considered
authorities either on fossil classification or on the issues about whether highly similar fossil
faunas and floras could be expected to be present at very different times.
Dissolving Rationality 213
41. In an illuminating series of diagrams, Rudwick analyses both the variations on the
basic themes (figures 15.2, 15.3, 15.4) and the trajectories followed by the participants
(figure 15.5). Austen and William Buckland emerge as the earliest adherents of something
like the eventual consensus.
42. Here I offer a ludicrous abbreviation of an extremely complicated story. Rudwick
makes it very clear how Murchison sometimes experienced apparently insuperable diffi-
culties both with the Continental strata and with his ally Sedgwick. Interestingly, many of
the geologists involved either were moving toward the final consensus or had adopted
Murchison's Devonian interpretation in 1839, on the basis of the analysis of rocks in the
Rhineland. See Rudwick's figure 15.5 and chapters 11-13.
214 The Advancement of Science
43. As Rudwick makes very clear, the hypothesis is credited to Murchison not because
he was the first to think along these lines, but because he pursued it more energetically
and with more resources (money and time) than anyone else.
Dissolving Rationality 215
44. I am greatly indebted to Martin Rudwick for helping me recognize this aspect of
the case.
216 The Advancement of Science
found in much older rocks. His efforts involve him in skirmishes with Mur-
chisonians, who argue that the interpretations of strata are suspect in these
cases. Murchison and his allies offer various hypotheses about where in Devon
the unconformity is to be found, but De La Beche and his supporters win
the resulting skirmishes. Neither side can be accused of undergoing inferior
cognitive processes, but, by the same token, neither side makes any headway
with its major problem.
Murchison's Continental ventures enlarge the scope of the controversy by
bringing into contention a position that initially appears impossible. To defend
the Devonian system it is necessary to show kinship between the North Devon
Greywacke and the Old Red Sandstone. Murchison's attack on the Rhineland
(with its many vicissitudes) goes some way in this direction, but, as Murchison
himself recognized, the case was clinched by the co-presence of the Old Red
fish with the Devon shells in the Russian strata. "It was the Russian evidence
of Old Red Sandstone fish in close association with Devonshire shells, Mur-
chison recalled, that had 'entirely dispelled any doubts' about the Devonian
interpretation" (Rudwick 1985 395).
I claim that, to a first approximation, the acceptance of the Devonian
system is based upon the following reasoning, articulated in different ways
by participants who focus on different details:
1. There are three available interpretations of the Devon strata: (a) they
are ancient, and the plant fossils at the top of the Greywacke antedate
the Old Red Sandstone; (b) the topmost strata, including those bearing
the plant fossils, are Carboniferous, while lower strata antedate the
Old Red Sandstone; (c) they form a continuous sequence from lower
strata contemporaneous with the Old Red Sandstone up to Carbonif-
erous strata at the top.
2. If (a) were correct, then it would have to be possible for similar fossils
to be found in rocks of radically different ages; other apparent examples
in which this occurs have proved to be based on errors in interpreting
the strata; hence, there is no independent reason to believe that the
Devon strata contradict the generalization that highly similar fossil
faunas or floras are not found in strata that are widely separated
temporally.45
3. If (b) were true, there would have to be an unconformity between the
Carboniferous strata at the top of the North Devon Greywacke and
the ancient rocks below, but sustained efforts to find this unconformity
have failed.46
45. As the controversy evolves, the kinship of the North Devon plants and Carbonif-
erous fossils becomes ever clearer. By the same token, the disparity between the fossils at
the top of the Devon Greywacke and the fossils of Murchison's Silurian system is more
and more evident. In consequence, De La Beche's problem of defending the ancient age
of the topmost North Devon strata is further aggravated.
46. Here, too, the problem becomes more recalcitrant through the controversy, as
Murchison and his allies explore the most promising places for discovering an unconformity,
with no success.
Dissolving Rationality 217
Greywacke and the Old Red within Britain. For the elite, the kinship of the
strata in Russia was enough, because their wider knowledge of rock types
and fossils generated no expectation that contemporaneous rock types need
exhibit their similarities within some arbitrarily designated area. Thus Williams
failed to appreciate both (2) and (4). Weaver, on the other hand, seems to
have been so devoted to making up for his Irish mistake that he was deter-
mined to seek out the elusive unconformity no matter how long or fruitless
the search might be.49
My view of Weaver and Williams can easily be illustrated with an analogy.
Imagine that a group of people are judging the finish of a race. By the finish
line, in prime position to register judgment, stands a cluster of individuals
who have wide experience in calling the result. Up in the grandstands are
two others, neither with an unoccluded view, who, if they stand on tiptoe
and crane their necks, can just see the finish line. Neither has ever judged a
race before. The runners cross, and everyone announces a verdict. All those
by the finish line agree. The two in the grandstands disagree with the consensus
and with each other. No devotion to "symmetrical explanation of belief"
should lead us to overlook the circumstances that render one set of verdicts
more likely to be correct and that cast doubt on the other. So it should
be with cognitive appraisal generally, and with Williams and Weaver
in particular.
There is no imputation of irrationality here. Weaver, perhaps, is the victim
of an idee fixe, and Williams, through the circumstances of his vocation and
his training, is in no position to recognize the full cogency of the elite rea-
soning. But dissolving the notion of rationality does not mean abandoning
cognitive appraisal. I claim that the framework of this chapter enables us to
develop Rudwick's account of the great Devonian controversy and to see
how, in this instance too, the compromise model applies. Out of socially
guided interactions with nature, there emerge superior ways of forming be-
liefs, and, because these were incorporated into the psychological lives of
virtually all the participants, the great Devonian controversy was brought to
a cognitively appropriate close. Society, nature, and sound individual rea-
soning combined to drive the social learning machine to a new success.
49. Speculation about Williams is easier because Rudwick provides more information
about his ultimate reactions to the consensus. However, in neither instance would I want
to claim that my psychological accounts are correct. They are simply intended as articu-
lations of the idea that both men underwent processes that were less likely to promote
cognitive progress than those that moved the geological elite. Explaining human belief
requires us to take seriously the causal processes that generate and sustain belief, without
assuming (on a priori grounds?) that all such processes will be equally likely to disclose
the truth about nature.
7
The Experimental Philosophy
219
220 The Advancement of Science
Since the early 1960s, dynamic models of human knowledge have been
popular among philosophers of science. Even before that, however, Popper
had made the issue of the growth of knowledge central to his treatment of
epistemological questions.3 Moreover, many of the insights of theories of
confirmation advanced within the static model can be reformulated within a
dynamic model, for example that offered by contemporary Bayesians.4 While
these articulations of a dynamic model are less artificial than the static ap-
proaches, which presuppose the availability to the subject of a vast corpus of
"evidence," they are still at an enormous remove from the practice of science.
Since the publication of (Kuhn 1962), several philosophers have broadened
the conception of cognitive state to include other types of commitments. So,
for example, both Laudan and Shapere think of the growth of scientific knowl-
edge in terms of the evolution of language, methodology, aims, and instru-
ments. This broadening of the notion of cognitive state is continued in my
introduction of individual and consensus practices in Chapter 3. The aim, of
course, is to add realism to philosophical accounts of science, accounts that
have often been criticized for their apparent lack of connection with the
phenomena.
Introducing the notion of an individual practice to capture the complexities
of scientists' commitments is only one facet of my departure from previous
models of scientific knowledge. The notion of experience as the source of
evidence, celebrated in the static model, is transformed in prior versions of
the dynamic model by supposing that experience is the motor of scientific
change. In my account, there are two processes that spark the modification
of individual practices: encounters with nature and conversations with peers.5
The second of these is ignored in traditional models of knowledge: perhaps
because it is taken for granted that the effects of social interactions are some-
how reducible to interactions with asocial nature. By contrast, those who offer
the sharpest challenges to philosophical accounts of scientific knowledge fre-
quently contend that this neglected source of change in belief is paramount
in science, and that the role of "encounters with nature" is insignificant.6 By
of rationality that I proposed in the previous chapter, it will hardly come as a surprise if I
claim that these issues dissolve under disambiguation of the kindred notion of justification.
3. Of course, in Popper's version the concept of justification gives way to a different
notion of appraisal: transitions are never justified, although they might be scientific or (in
Lakatos's formulation) intellectually honest (Lakatos 1970).
4. See, for example, (Jeffrey 1983, Howson and Urbach 1989, Earman 1992). Non-
Bayesian approaches that share the commitment to thinking in terms of assignments of
probabilities to statements can also articulate a dynamic model of knowledge. See (Levi
1982).
5. It is worth reemphasizing that encounters with nature include those occasions on
which scientists reason by themselves. Thus, I want to allow for the creative process of
reasoning to a new idea, which resolves some of the antecedent tensions in practice. As I
noted in Chapter 3, such tensions are always present. The account of mathematical knowl-
edge offered in my (1983a) saw them as sources of scientific change and as dominant sources
of mathematical change.
6. Although they would not put the point in this way, I think that Harry Collins (1985)
and Andrew Pickering (1984) suggest a dynamic model of human "knowledge" in which
The Experimental Philosophy 221
modifications of the complex states of individual scientists are caused by social interactions
among them. The model can probably be traced to (Bloor 1974). Perhaps the best way to
contrast it with the usual philosophical approaches is to see that whereas philosophers
typically take the social input from others to be ultimately reducible to the impact of nature,
the sociologists whom I have mentioned regard encounters with nature as so thoroughly
mediated by society that they do not provide any independent constraint on the cognitive
states of scientists.
222 The Advancement of Science
2. Observation
Start with a relatively simple situation. A behavioral biologist is observing a
baboon troop. Over a period of several hours he records the episodes in which
one of the animals grooms another, carefully noting the names of the animals
(who groomed whom) and the time interval through which grooming oc-
curred. Each entry in the notebook records the perceptual acquisition of a
belief. Focus on any one. The observer is initially scanning the troop. He
sees the male he calls "Caliban" approach the female he calls "Miranda."
There is a sequence of facial expressions and gestures, at the end of which
Caliban crouches behind Miranda and plucks at her fur. Our biologist presses
a button on his stopwatch and quietly moves to a position from which he can
gain a better angle on the interaction. After a few minutes, Miranda shrugs
and moves away. Another button on the stopwatch is pressed, and the bi-
ologist writes in the notebook, "CalibanMiranda, 6:43." That notation
serves as an extension of declarative memory, something from which the
biologist can later retrieve the belief that Caliban groomed Miranda for a
period of six minutes and forty-three seconds.
The "observation reports" in the notebook are highly unlikely ever to
figure in public scientific discussion. Assume that the biologist's principal
interest is in differences in times spent grooming between relatives and non-
relatives. The "data table" of the published report will divide up the inter-
actions by applying background views about the relationships among the
animals and then sum the time periods in various divisions. Caliban and
Miranda may vanish from the scene; the individual episode of the last par-
agraph will almost certainly do so.
Assembling the published "data" typically relies on inference as well as
observation, and almost always involves trusting the credibility of others.
Simple observation rarely warrants, by itself, any modification of individual
practice: justifications of the acceptance and rejection of statements, even at
the level of "empirical data," are usually complex (see Bogen and Woodward
1988, Galison 1987). But even baboon-watching is more subtle than it might
initially appear.
Consider the pitfalls that attend the performance of the imaginary ob-
server. Perhaps the participants have been misidentified. Perhaps the inter-
action between them has been misclassified. Perhaps the stopwatch is not
functioning properly. Perhaps the presence or the movement of the scientist
disrupts the normal activity of the troop. (This last point is important, for
the "data" eventually published will purport to characterize the behavior of
baboons under undisturbed conditions. Indeed, the scientist may include a
defense of the claim that his presence did not distort the pattern of behavior.)
These are not instances of hyperbolic doubt, but live possibilities for an
unskilled or unlucky observer to go astray. Typically, skilled observers do
not proceed by running through a checklist, assuring themselves that the
possible pitfalls are absent. They operate by habit.
The Experimental Philosophy 223
The skilled observer has abilities to reidentify and classify that the neo-
phyte lacks. While the newly arrived graduate student must painfully look
for clues to a baboon's identity, long experience of watching the troop enables
the biologist to pick out Caliban at once. Moreover, the ordinary, unself-
conscious habits of observation can be inhibited if the environment signals
the threat of a pitfall. When the stopwatch has recently fallen into the wa-
terhole, the biologist will take pains to check its working before relying on
it in further observations. And, as I intimated in describing the original sit-
uation, when he moves to improve his view of an interaction, the biologist
will consciously try to avoid startling the baboons.
The trained biologist differs from the neophyte in two ways, using different
concepts for reporting what is seen ("Caliban groomed Miranda" versus "A
big baboon sat down next to a smaller one") and having different propensities
for belief formation. The newly arrived research student is intermediate,
possessing the concepts that the biologist employs but applying them with
greater self-consciousness and less fluency. Even the apparently simple ob-
servational episode with which we began is shaped by the past, in the training
of the skilled observer and in the history of the consensus that stands behind
it. The approach to progress that I developed and defended in Chapters 4
and 5 is optimistic about what has been accomplished. The trained observer
has learned to see well: using more adequate concepts in reporting what is
seen, making discriminations that the untrained observer cannot make, ac-
quiring propensities that work both more efficiently and more reliably than
the explicitly clue-following processes of the research student.
This dependence on the past can be viewed more pessimistically.7 Since
the beliefs that people form depend not only on the stimuli they receive but
on their background cognitive stateson the languages they are able to use
and the propensities that have been stored through trainingit is possible
that two observers should receive the same stimuli and form quite different
beliefs. This is unworrying if one (and only one) of the observers can be
presented with stimuli that generate beliefs from which, by reasoning that
accords with shared canons,8 it is possible to arrive at the distinctive features
of the cognitive state of the other. In this case there is an epistemic asymmetry,
and the superiority of one of the ways of responding to the stimuli can be
defended. But if habits of observation and observational reporting sometimes
change without the possibility of justifying the modification, then there is
7. The pessimism is apparent in the writings of Kuhn and Feyerabend on the theory-
ladenness of observation, especially (Kuhn 1962/70 chapter X and Feyerabend 1975). Both
Kuhn and Feyerabend start from the insight that there can be variation in reporting of
observations that cannot be traced to any obvious malfunctioning of one of the observers
and conclude that there are possibilities of championing alternative doctrines that cannot
be discriminated through appeal to observation. As I argue in the text, the conclusion
depends on acceptance of a type of epistemic symmetry that goes beyond their initial insight.
8. I assume here that the shared canons of good reasoning, that is, that they are likely
to lead to the inculcation of true beliefs. Even though both scientists might fail to employ
optimal belief-modifying strategies, I shall also suppose that use of such strategies would
break the symmetry in the same fashion.
224 The Advancement of Science
room for concern that observation might be unreliable. Perhaps the "skills"
inculcated by training are not skills at all.
Let us examine this argument from epistemic symmetry more carefully.
Imagine that there are two observers with background cognitive states C1 and
C2, and that there is a stimulus s that induces in the first a belief that p1 and
in the second a belief that p2, where p1 and p2 are incompatible. There is a
condition of epistemic symmetry only if (i) there are no stimuli that could
induce a belief that p2 in a subject with background state C1 (ii) there are
no stimuli that could induce a belief that pl in a subject with background
state C2, (iii) there are no processes of accepted good inference available to
a subject with background state Cl that would warrant belief that p2, (iv)
there are no processes of accepted good inference available to a subject with
background state C2 that would warrant belief that p1, (v) there are no pro-
cesses of accepted good inference available to a subject with background state
C1 that would modify the cognitive state in such a way that there would be
stimuli inducing belief that p2, and (vi) there are no processes of accepted
good inference available to a subject with background state C2 that would
modify the cognitive state in such a way that there would be stimuli inducing
belief that p1.9 It is now claimed that, in the history of science (specifically
in the history of scientific controversy), there are rival observers with different
background states who are led to incompatible beliefs by the same stimulus
and with respect to whom a condition of epistemic symmetry prevails. One
of the rival positions and the associated concepts and propensities for obser-
vation survive in our contemporary science. But we cannot claim that this
survival marks an advance in ability to observe or that the "skills" are reliable
generators of perceptual knowledge. For at least one of the systems of concepts
and propensities must generate error (since they yield incompatible beliefs),
and, by the hypothesis of epistemic symmetry, there is no way of telling
which.
Making this skeptical argument work is far more difficult than appeals to
the theory-ladenness of observation typically suggest. It is not enough simply
to declare that rival scientists would have been inclined to respond to the
same stimulus by asserting incompatible statements: Priestley announces that
the stuff in the flask is dephlogisticated air, Lavoisier contends that it is
oxygen, Galileo maintains that a point of light in the sky is a translunar comet,
his Aristotelian opponent counters with the claim that it is a sublunary ex-
halation. What must be shown is that the conditions for epistemic symmetry
are met, and, since the psychological states of Aristotelians, Galileo, Lavoi-
sier, and Priestley are not easily accessible, the skeptic has to proceed indi-
rectly by using information about the contemporary performances of
Figure 7.1 The Muller-Lyer Illusion. Although the line on the right looks longer
than that on the left, both lines have the same length.
10. This point has been appreciated in some recent, sensitive studies of observation
and scientists' reporting of their observations. See, for example, (Shapere 1982, Pinch
1985), both of whom recognize that, at moments of controversy, scientists might respond
to stimuli using more minimaland unproblematiccategories. Shapere and I share the
view that the use of more recondite concepts can be understood in terms of reliable processes
of reasoning; Pinch prefers to explain such usage by appealing to social processes of
negotiation.
The Experimental Philosophy 227
15. Numerous reports in the writings of Galileo and his contemporaries attest to the
success of terrestrial demonstrations of the veracity of the telescope: Galileo describes how
the elder statesmen of Venice puffed up the stairs to turn the telescope on the sea, the
members of the Lincean Academy tell of using the telescope to read the inscriptions on
distant buildings, and even the critical Martin Horky concedes that "below it works won-
derfully." See (Galileo/Favaro III, Drake 1978, Van Helden 1989). Feyerabend is very
clear that there was no problem in checking the reliability of the telescope on earth (1975
107-108).
I should note that Ian Hacking's extended discussion of observation and the establish-
ment of observational skills (1973 chapters 9-11) brings out in an exemplary way the kinds
of strategies that historical figureslike Galileohave employed. My own treatment is
indebted to Hacking's, although I place far less emphasis on the role of manipulation. My
account is also akin to Richard Miller's discussions of the extension of observation in
his (1987), although I am concerned to emphasize the general structure of the ways in
which skills are established and thus dissent from his suggestion that confirmation is merely
local. For explicit contrast, see (Kitcher 1992).
16. For example, Galileo introduced the technique of stopping down a lenspossibly
as the result of his own eye trouble, and his practical discovery that he could improve his
vision by partially closing his eyes (Drake 1978 148). As we shall see, Galileo also made
better telescopes than those who tried to emulate him, and an important part of his campaign
for the telescope depended on his devoting large amounts of time, energy, and money to
making and circulating instruments with lenses ground to his specifications.
The Experimental Philosophy 229
and Miranda close at handto contexts in which they have not been inde-
pendently vindicated. But background practice provides no grounds on which
to distinguish (say) episodes of identification that have been carried out up
to now (taken to be reliable) and episodes of identification in the future
(assumed not to be reliable). By contrast, the practices of Galileo's contem-
porary Aristotelians generated principled reasons for thinking that Galileo's
instrument might only work well on the earth.
Now the obvious and direct way of extending Galileo's demonstrations to
undercut this objection would be to show some superlunary phenomenon that
could not be appreciated with the naked eye and then put oneself in a position
to check it directly through naked-eye observations. Lacking spaceships, Gali-
leo was unable to do this.17 Forced to proceed in a more roundabout way,
Galileo responded to the Aristotelian objection by combining three strategies:
first, he appealed to tempered naked-eye observations to argue that the
celestial-terrestrial distinction is misconceived; second, by distributing his own
telescopes and providing specifications of how to use them and pictures of
the phenomena he claimed to have seen, Galileo could assemble a body of
tempered reports of observations, produced by observers working in different
places, that would testify to the regularity of his phenomena; third, he could
connect his telescopic observations to phenomena that are barely visible with
the naked eye.
The first strategy had already been implemented in the debate over the
placing of the nova of 1604. Committed Aristotelians attempted to resist the
claim that a new star had appeared in the allegedly changeless heavens. Faced
with calculations of distance based on parallax which seemed to place the
nova well beyond the sphere of the moon, they tried, in various ways, to
explain away either the apparent distance or the apparent novelty.18 Galileo
continued to argue, from the early 1600s through into the 1630s, that the
celestial/terrestrial distinction was untenable, not only because of the presence
of comets and novae in the heavens but because of internal difficulties in the
Aristotelian conception.19
17. Thus van Fraassen's claims about observability, while they capture the importance
of checking putative observational skills against already accepted observational abilities,
are quite detached from the ways in which scientists actually convince themselves and others
that they have legitimately extended the senses (Van Fraassen 1980 16). In principle,
however, Galileo might have been lucky enough to observe a comet approaching the Earth
so that he could have used the telescope to predict its appearance. This would be analogous
to his demonstrations with incoming ships.
18. One noteworthy effort was that of Ludovico delle Colombe in 1605, who suggested
that the nova was an old, distant star that appears to us magnified by the crystalline spheres.
In a devastating reply, allegedly "The Considerations of Alimberto Mauri," an unknown
author (or authors) attacked Colombe by criticizing arguments for the incorruptibility of
the heavens, by showing that Colombe's account gave a ludicrously bad estimate of the
time of visibility of the nova and demonstrating how it was incompatible with everyone's
astronomical theories. On the basis of both style and content, Drake speculates that Galileo
may have had a hand in this work (1978 118).
19. The former types of argument culminate in Day Three of the Dialogue, the latter
in Day One.
230 The Advancement of Science
The third strategy, that of trying to show that telescopic observations are
continuous with naked-eye observations, could be pursued in showing how
the telescope apparently added greater detail to a picture whose outlines were
known in advance. Thus Galileo's drawings of the moon and of two con-
stellations (Orion and the Pleiades) reveal familiar featuresthe "darkish
and rather large" spots on the surface of the moon that have been seen since
antiquity, the previously identified members of the constellations in their well-
known positionsbut embed these familiar features in a richer context. Po-
tentially available to Galileo was the tactic of showing how successively more
powerful telescopes would disclose more and more of the detail. Since his
own construction of telescopes began with a low-magnification instrument, it
is quite possible that his own conviction that he was seeing more and more
of the fine structure of the heavens rested on a sequence of inspections of
the same astronomical phenomenon using progressively more powerful tele-
scopes, but I do not know whether he ever appealed to any such sequence
in defending the reliability of the telescope.
By far the most prominent strategy in Galileo's attempt to establish his
observational virtuosity centered on his campaign to help others to see the
four "Medicean planets" (the moons of Jupiter). The reasons for this are not
hard to discern: Galileo's primary goal in 1610 was not simply to establish
the fact that the telescope could give accurate information about the heavens
or to employ the telescope to vindicate Copernicanism and debunk
Aristotelianism20; his aim was to show that he had made a specific discovery,
worthy of being rewarded with a position at the Florentine court.21 Because
he made the existence of the Medicean planets central to his own claims, his
opponents, particularly Horky and Sizzi, joined debate with him on precisely
this issue.
How can anyone simultaneously campaign for a new phenomenon and for
the reliability of the instrument that is used to uncover it? We have already
looked briefly at ancillary arguments that would undercut various kinds of
Aristotelian challenge. The centerpiece of Galileo's case, however, is the
possibility of obtaining, from astronomical observers working in different
parts of Europe, a systematic collection of sightings that show both (a) agree-
ment with one another on locations assigned on the same night and (b) a
sequence of assigned locations that concur with those represented in the
Sidereus Nuncius in showing apparent regular rotation about Jupiter. By 1612
Galileo achieved just this, as observers from around Europe jointly contrib-
uted to a corpus of claims satisfying (a) and (b). In consequence, many of
20. Here Feyerabend goes astray in suggesting that Galileo's Copernicanism played a
central role in his defense of the telescope. (Westfall 1985) shows how Galileo only turns
to a Copernican program of astronomical observation after he has successfully defended
the telescope.
21. This is evident not only from the dedicatory strategy of Sidereus Nuncius, but from
the actions Galileo took to ensure that the Medici in particular would have ample confidence
in his discovery of the moons of Jupiter. See the penetrating analyses by R. S. Westfall
(1985) and Mario Biagioli (1990).
The Experimental Philosophy 231
22. One particularly important group whom Galileo convinced were the Roman Jesuits.
Late in 1610, Jesuit observers in Rome were able to report that they had seen the Medicean
planets, and, in 1611, at the instigation of Cardinal Bellarmine, they produced an official
report on Galileo's findings. As I shall indicate, agreement with Galileo often exists at the
level of tempered observations, and Jesuits such as Clavius explore alternative explanations
of these tempered observations (see Drake 1978 165). But this is already to recognize that
the telescope is reliable about some aspects of the heavens (even if, from a traditionalist
perspective, Galileo is inclined to overinterpret its deliverances).
232 The Advancement of Science
with me in a house observed the heavens on the same night of 25 April. But
all acknowledged that the instrument deceived. And Galileo became silent,
and on the twenty-sixth, a Monday, dejected, he took his leave from Mr.
Magini very early in the morning. (Van Helden 1989 92-93)
Magini's own comments (also expressed to Kepler) were more guarded. He
reported that "nobody has seen the new planets distinctly" and doubted
whether Galileo's discoveries would stand up.
Galileo's dejection seems only to have been temporary. Throughout the
second half of 1610 he distributed high-quality telescopes to prominent centers
of culture in Europe, providing detailed instructions for the observation of
the Medicean planets. In the light of the growing number of reports that the
combination of Galileo's instruments, his instructions, and his diagrams for
showing the various arrangements of the moons (essentially indicating what
the phenomena were supposed to look like!), even those who had initially
had difficulty in making observations with the telescope changed their minds.
After Horky published his attack on Galileo, Kepler broke relations with
him, Magini chased him out of the house, and Roffeni, on behalf of the
Bolognese faculty, Magini, and himself, published an apologetic reply to
Horky's tract.
By late 1610, those who were still undecided about the reliability of the
telescope had two options: either they could view those who reported the
positions of the Medicean planets as duped in some systematic way or they
could suppose that failure to see the planets resulted from the absence of
observational skill. The latter, of course, conceded to Galileo everything
he claimed. The former required acceptance of minimal reports of obser-
vation about the appearance of points of light in specific places at specific
times and shifted the issue to find an explanation of those appearances in
terms of the "deceptions of the telescope." With no such explanations
available, it is hardly surprising that thoughtful, conscientious people came
to adopt Galileo's interpretation and accept the deliverances of the telescope
at face value.
Feyerabend's skeptical argument fails to take into account both the
coordination of the observations reported by different people and the or-
derliness of the successive observations of single observers. Both are ex-
pressed in discussions of the period of the satellites. He also neglects
Galileo's technological improvements: by the end of 1610 his telescopes
were better than they were at the beginning, and it was thus easier for him
to obtain agreement from others. However, at no stage did Galileo have
an adequate theoretical account of his telescope, and in this Feyerabend
finds a crucial flaw.
According to Galileo's own account, his original construction of the tele-
scope was based on his understanding of "the science of refraction" (Van
Helden 1989 37). Yet no satisfactory explanation of the properties of lenses
had been published by 1610, and, as Feyerabend points out, Galileo's "knowl-
edge of optics was inferior by far to that of Kepler" (1975 105). But Galileo,
as nobody else in the period, combined a knowledge of optics with a rich
The Experimental Philosophy 233
repertoire of "craft knowledge." Trial and error, guided by both his practical
experience with lenses and his acquaintance with mathematical optics, led
him to a series of telescopic improvements (see Van Helden 1977 18, 26-27).
No doubt a full theory of the telescope would have been very helpful to him,
but it was not needed for devising his instrument, for improving it, or for
vindicating its accuracy. Ignoring the multiplication of systematic observations
of the heavens and focusing on the most celebrated incident in which dem-
onstration of the telescope failed, Feyerabend concludes that only the pro-
vision of a theory of the telescope would have underwritten Galileo's claims.
As I hope to have shown, Galileo used quite different means to display his
observational virtuosity, means that cannot be written off as propaganda.
I conclude that we can recognize the dependence of observation and of
observational reporting on background cognitive states without abandoning
the thesis that the propensities acquired by specialists are observational skills.
Trained professionals can demonstrate their virtuosity in ways that are ap-
preciable by the laity, and so turn back the challenge that their alleged skill
is merely masquerade.23
3. Inductive Generalization
Scientific research, like everyday life, would be impossible without the activity
of generalization. The evaluation of observational techniques, of others' cred-
ibility, requires using the past record as a guide to future performance.24 Even
more obviously, to anticipate the future course of experience, to guard against
danger, and to exploit opportunity, people have to use the information avail-
ableand what is available is, of course, information about the past, about
what has been observed. The issue is not whether the things we have checked
should guide our beliefs about what remains unchecked, but how we should
be guided.
Let us begin with a common inductive situation. Imagine that scientists
(or people pursuing practical projects) have isolated a set of entities, the A's,
23. This is not to say that there will never be difficult cases. For it is quite possible that
the kinds of strategies I have described for checking a new instrument (or those that Hacking
discusses in his (1983)) would be unavailable because of the absence of any overlap with
the deliverances of other, unproblematic, modes of observation. If an instrument is designed
to do a very specific job which is discontinuous with the functions performed by other
pieces of established technology, there may simply be no basis from which to judge its
reliability. Perhaps this is the situation with Weber's apparatus for detecting gravity waves,
described with great force by Collins in chapter 3 of his (1985). However, for a significant
class of complex instruments, it is possible to develop defenses of their reliability along the
lines indicated here. See the lucid account offered by Peter Galison in his (1987).
24. Thus a full account of the establishment of an observational skillsuch as Galileo's
defense of the use of the telescope in astronomical observationdepends on recognizing
the effective strategies of generalization. My discussion in the last section thus proceeds at
an intuitive level, relying on everyday views about how to generalize without identifying
them precisely.
234 The Advancement of Science
and that the question "How do the A's exemplify D(B)?" is significant for
themwhere B is some determinate property (for example, being blue) and
D(B) is the corresponding determinable (in this instance, color). For problems
of this form the possible answers can be represented as n-tuples (P1, . . . , pn)
such that the pi sum to one.25 The number n is the number of determinate
forms of the determinable D(B). Cases in which one of the pi is one and the
rest zero are pure generalizations, in which the answer is "All A's are C"
(where C is a determinate form of D(B)). Mixed generalizations are answers
that assign nonzero frequencies to two or more of the determinate forms-
as, for example, when half of the A's are declared to be B and half C. An
inductive problem situation is constituted by the initial demarcation of the
set over which generalization is sought (the A's), the determinable with respect
to which generalization is sought (D(B))these, as we have just seen, fix the
set of possible answersand the sample knowledge. The latter consists in the
specification of the frequencies with which the determinate forms of D(B)
have been found in those A's that have been observedthe members of
O(A)which can be represented as an n-tuple (p01,. . . , p0n), together with
statements about some of the characteristics of O(A). Since O(A) is a finite
proper subset of A there are always two or more possible answers that are
consistent with the observed frequencies. Adding an answer to the set of
accepted statements always involves the risk of error, even if the specification
of the observed frequencies is correct.
I suggested earlier that the important question about inductive generali-
zation is how to let past experience guide our anticipations of the future. We
can now see that this question divides into two parts: When should we accept
some one of the possible answers? Which answer should we accept? Plainly,
the need for relief from ignorance may be more or less acute. In everyday
life, we sometimes have no choice but to select some one of the answers.
Similar situations arise within the practice of science, particularly when sci-
entists have to assess the reliability of techniques, instruments, and potential
informants. In other instances there is less urgency, and the choice for a
scientist is initially whether to pick some answer now or whether to seek more
information (typically by expanding O(A)). If the first option is chosen, there
is then a decision about which answer to pick; if the second, the next step is
to decide how to obtain more information.
Inductive propensities are far more complex than is usually supposed.
Even for the simple type of situation with which we are presently concerned,
I shall conceive of the inductive propensity as involving (i) the framing of the
problem through the specification of A and D(B), (ii) a disposition to accept
some answer to the question "How do the A's exemplify D(B)?" if and only
25. Strictly speaking, this holds for discrete cases in which there are a finite number of
determinate properties corresponding to the determinable D(B). Some problems involving
the investigation of magnitudes can be conceived as either discrete or continuous. Thus a
scientist might inquire how the entities in a set are assigned values of the magnitude in
question (continuous form) or which of a finite set of functional dependencies is realized
within this set (discrete form).
The Experimental Philosophy 235
if the subject believes that O(A) meets a condition R, and (iii) a function /
that maps the frequencies of the determinate properties within O(A),
(p01, . . . , p0n), onto the answer, (P1, , pn), that is accepted when the subject
believes that O(A) meets R. The inductive propensities in which I am inter-
ested have to satisfy a condition of psychological realizability: people have
to be disposed to identify the A's, disposed to pick out D(B) and its deter-
minate forms, able to judge whether a set meets R, and able to give a general
specification of f. Methodologists should try to explain what realizable dis-
positions to pick out A, D(B), R, and f will work well in combination.
The most obvious question about inductive generalization seems to be
how to choose the function /. But, I suggest, we can narrow our focus by
concentrating on the special case in which / is identity. By doing so, we
generate a family of inductive propensities that correspond to the straight
rule ("Infer that the frequency with which a trait is found in a sample is that
with which it is found in the entire population"). Members of the family differ
with respect to the ways in which they frame inductive problems (choices
about combinations of A and D(B)) and their standards for judging that a
sample is representative (selection of R).
An inductive propensity, then, consists in a disposition to generalize (fol-
low the straight rule) with respect to certain ways of classifying entities and
picking out their properties and given certain beliefs about the observed
instances. Since Nelson Goodman's seminal (1955), it has been obvious that
inductive generalization cannot be completely understood without taking into
account the types of classification involved. For any set and determinable,
we can consider the subsets observation of whose members would suffice to
prompt us to generalize: intuitively, these subsets comprise what we take to
be representative samples. If there are no such proper subsets, then the pair
A, D(B) is (in Goodman's term) unprojectible. We cannot expect to learn
the answer to the question, How do the A's exemplify D(B)? by looking at
instances. Conversely, well-posed inductive problems are those that do allow
for representative samples that are proper subsets of the set over which we
intend to generalize.
The normal form of an inductive generalization is as follows:
and to apply the straight rule when the sample is judged to meet a certain
condition. Let us represent this as (F, R), where F selects the appropriate
(A, D(B)) pairs and R is the condition on the sample for the activation of
the propensity for straight generalization.26
The methodologist's goal is to discover which pairs (F, R) are good ones.27
Our next task is to articulate an appropriate standard of goodness. There are
various ways to proceed. The simplest is to consider a range of situations in
which an inductive propensity might be applied. In some of these, the con-
ditions for the activation of straight generalization will not be met, and a
subject who employs the inductive propensity will not perform any inductive
inference. Call these the undecided cases, and the rest the decided cases.
Within the decided cases, the propensity sometimes yields true conclusions
and sometimes false conclusions. The former are the positive cases and the
latter the negative cases. Let the ratio of the decided cases to the undecided
cases be the relief index, and the ratio of the positive cases to the negative
cases be the truth ratio. Ideally we would like propensities with high relief
indices and high truth ratios. However, it is not hard to see that the two
desiderata pull in opposite directions: the more likely a propensity is to pro-
vide relief from agnosticism, the less cautious it is likely to be, and the
abandonment of caution typically lowers the truth ratio. Thus one problem
that has to be faced is how to make an appropriate trade-off between the two
virtues. Another difficulty in articulating a standard is to decide what the total
range of situations we envisage ought to be.
Just as the set of inductive propensities we consider is restricted to those
that are psychologically realizable, so too the range of situations over which
those propensities are evaluated should encompass only situations that arise
in the actual world. The methodologist hopes to advise human beings inves-
tigating our world. However, to specify appropriate types of situations we
have to turn to history and to science. History tells us what kinds of problems
scientists and others have had to face in the past; science informs us about
the solutions.28 Drawing a representative sample, we evaluate inductive pro-
pensities by computing their relief indices and their truth ratios. So our present
perspective on the world and the investigation of it is employed to judge the
success of various ways in which our predecessors might have done their
generalizing.
My talk of a representative sample is a fudge. There is no obvious way
to select from the past some set of problem situations that will encompass
the puzzles that typical people have to overcome. Human experience is suf-
ficiently diverse that it might be better to assess inductive propensities by
looking at their performance with respect to many sets, each of the normal
sets of problem situations. Intuitively, we eliminate as pathological certain
kinds of "inductive experiences"for example, those in which a subject al-
ways encounters problems whose correct solution is indicated by the first
object sampled and those in which long runs of instances of one kind are
eventually followed by even longer reversals. The remaining inductive ex-
periences, those consisting of sequences of problems that we think a subject
might encounter, count as normal sets of problem situations. Different in-
ductive propensities might, of course, fare better or worse on different normal
sets. What we want to discover is which propensities do well across the board,
which do reasonably well in all cases and extremely well in some, and so
forth. I do not pretend that there is any single correct way to articulate the
standard.
Even the simplest approach to setting standards for inductive propensities
encounters two sources of indeterminacy, one surrounding the trade-off be-
tween relief from agnosticism and deliverance of truth, and one concerned
with delimiting the sets of problem situations with respect to which the as-
sessment is to be conducted. Matters are further complicated if we explore
various refinements: (a) taking into account not only whether or not the
conclusion coincides with the correct answer but also its distance from the
correct answer, (b) considering the costs of various kinds of errors (so that
incautious inductive propensities might be heavily penalized for delivering
incorrect conclusions about certain kinds of problems), (c) recognizing the
practical or cognitive significance of certain types of questions (so that highly
cautious inductive propensities might be heavily penalized for not delivering
answers). Taking these factors into account allows yet more indeterminacy,
so that the range of appropriate standards for evaluating inductive propensities
is further increased. In accordance with the discussions of Chapters 3 and 6,
I draw the moral that the notion of a single "correct" or "rational" inductive
propensity is a myth. That is not to declare that all inductive propensities are
equally good. Some may be classed as definitely inferior whatever standard
we choose.
4. Eliminative Induction
I want to defend an old-fashioned idea about human inductive propensities
or, more exactly, about the human inductive propensities that are activated
in scientific inferences. According to this idea, induction proceeds through
the elimination of alternatives. The main task of this section will be to show
how to characterize an inductive propensity that works eliminatively and how
to understand certain types of scientific inference in terms of the activation
of this propensity. In later sections, I shall try to show that this propensity
can be defended against some powerful objections.
238 The Advancement of Science
The motivating idea behind the eliminative propensity is that the prior
state of individual practice is used to select appropriate candidates for in-
ductive generalization and to construct a space of alternative possibilities.
Thus, for the question, How do the .A's exemplify D(B)?, to present an
inductive problem for subjects with the eliminative propensity, those subjects
must accept the statement that each A has some determinate form of D(B)
and they must use a language in which the set A is recognized as projectible
with respect to the determinate forms of D(B).29 The space of possibilities
that they construct depends on the accepted explanatory schemata, specifically
on views about how the determinate forms of D(B) depend on other properties
for objects of a kind K(A) to which the members of A all belong.
If there are several explanatory schemata that provide views about the
dependencies of the determinate forms of D(B) on other properties for dif-
ferent kinds that subsume A, then the space is constructed by considering the
narrowest of these kinds. The procedure for construction is as follows: Call
the properties on which the form of D(B) is explanatorily dependent across
K(A), the background conditions. Partition A into cells such that the members
of each cell exemplify a conjunctive condition each of whose conjuncts is a
background condition or its negation and which contains, for each background
condition, either that condition or its negation. (So, for example, if there are
two background conditions, X and Y, the partition of A has the four cells,
XY, XY, XY, X Y ) . The eliminative propensity relies on the idea
that the background conditions specify all the properties on which variation
in determinate forms of D(B) across K(A) depends. In cases in which the
subject accepts the statement that the presence of a form of D(B) in an A is
deterministic, the possible hypotheses are reduced to a finite number, all of
which claim that each cell is homogeneous for the form of D(B). With respect
to such situations, subjects with the eliminative propensity generate the only
hypothesis compatible with the findings in O(A) just in case O(A) is judged
to contain at least one member of each cell.
Consider the simplest type of situation in which the eliminative propensity
works. The problem is to explore the way in which A's exemplify D(B), and
we can imagine that m A's have been examined each of which possesses the
property B. The individual practice of the subject counts A as projectible
with respect to D(B) and recognizes k background conditions, C1,. . ., Ck.
Moreover, it is part of individual practice that the presence of determinate
forms of D(B) in members of A is deterministic. A is partitioned into 2k cells,
each of which is taken to be homogeneous with respect to the determinate
form of D(B). The subject generates the conclusion that all A's are B's just
in case she believes that O(A) contains at least one member from each of the
29. This condition requires that there be accepted generalizations linking sets similar
to A to determinate forms of D(E). The similarity is fixed by background views about
which sets of objects form kinds and which are subkinds of the same kind. The general
line I advocate is thus akin to that endorsed by Goodman in his own treatment of the "new
riddle of induction": projectibility stems from entrenchment and entrenchment comes from
the past history of successful projection (see chapter 4 of Goodman 1955).
The Experimental Philosophy 239
cells. Otherwise she seeks representatives of those cells that are not yet
represented in O(A). (Since there are 2k cells it seems that the cardinality of
O(A), m, has to be at least 2k if a conclusion is to be generated. I shall explore
this point later.)
The inference performed by the subject is deductive. Among the premises
is a claim that there are only a finite number of possibilities. Other premises
jointly imply that all but one of these possibilities is incompatible with the
premise that reports the properties of those A's that have already been ex-
amined. The conclusion, in the words of Sherlock Holmes, is that the re-
maining hypothesis, however improbable, must be the truth.30
So far I have only been attempting to characterize the eliminative pro-
pensity.31 That attempt raises some obvious questions: Where do the strong
premisesthe claims about explanatory dependence and so forthcome
from? Isn't inductive generalization needed to obtain them? How did the
whole process ever get going? Should an eliminative strategy allow for the
possibility that the currently specified background conditions are not com-
plete? These questions deserve answers, and I shall take them up later. First,
however, I want to show how the eliminative strategy functions in scientific
generalization.
Consider, first, the behavior of a chemist investigating reactions involving
a newly synthesized compound. The chemist hopes to arrive at generalizations
about the rates at which the reactions proceed, the energies released or
absorbed, and so forth. The investigation is framed by drawing on the ex-
planatory schemata of chemistry and singling out those properties (pressure,
ambient temperature, and so forth) on which variation in rates of reaction
and absorbed energies depend. Once the appropriate variables have been
"controlled for," the chemist is ready to announce whatever generalization
emerges unscathed.
Turn now to the philosopher's favorite, generalizations about the colors
of bird plumage. These are poor examples for strict inductive generalization,
because background biological theory suggests that there will be exceptions:
mutant alleles that direct the formation of abnormal pigments can easily arise.
However, there is a related form of generalization, prevalent throughout
biology, in which judgments about the overwhelming majority of members
of a species (or a higher taxon) are made on the basis of sampling ("Dic-
tyostelium discoideum aggregates in the presence of cAMP," "Impatiens grows
30. In his (1992), John Barman defends eliminative induction and explicitly makes the
link with Sherlock Holmes. My own treatment has been influenced by Earman's ideas about
scientific inference, although I believe that I am far more a partisan of eliminative induction
than he currently is.
31. Moreover, given the informal nature of some of my characterizations, I have only
succeeded in picking out a cluster of eliminative propensities. The exact details of how
background practice determines the set of pairs (F, R) have not been specified, for I have
not stated precise conditions on background practice for the set A to count as projectible
with respect to D(B), nor have I done more than indicate the way in which the spaces of
possibilities are constructed. (In accord with the pluralistic approach of the last section,
we might well allow for variation here.)
240 The Advancement of Science
32. Hence biological predicates picking out large sets of organisms are often taken to
be "almost-projectible" with respect to physiological, anatomical, behavioral and devel-
opmental properties, and this status shows up in virtue of the presence within the accepted
statements of practices of generalizations about the overwhelming majority of members of
the set. As Wesley Salmon reminded me, strict generalizations and probabilistic general-
izations (including "almost all" generalizations) have different logical properties: specifi-
cally, the latter do not admit of contraposition. This point can be used to evade the paradox
of the ravens, although, as I shall argue in the next section, my approach to induction
allows for a general solution of that paradox.
The Experimental Philosophy 241
33. Here there are also affinities to the ideas of Popper and Lakatos about "rigorous
testing." Andrew Wayne has pointed out to me that there are many similarities between
the arguments of this chapter and Lakatos's approach to scientific methodology. This seems
to me most evident in my approach to problems of underdetermination and in my view of
the shortcomings of Bayesianism.
34. To say that our primitive apparatus is relatively well designed is emphatically not
to assume that it delivers truths about regularities in the behavior of middle-sized physical
objects. See, for example, (Stich 1990) and the discussion in 10.3. The account of inductive
propensities I give here is analogous to that advanced for our primitive mathematical
knowledge in my (1983a).
242 The Advancement of Science
the primitive apparatus stands behind our primitive scientific practices. With
those practices in place, the eliminative propensity can be activated, and the
use of that propensity (together with other types of inference to be considered
in Section 7) allows for the modification of practice, the revision of the prim-
itive categorizations and views of dependence.
5. Positive Instances
I want to approach other concerns about eliminative induction indirectly, by
considering some familiar issues about confirmation of hypotheses. I shall
start by using the well-known paradox of the ravens to introduce issues about
the value of positive instances.
In a landmark article (Hempel 1945), C. G. Hempel pointed out that the
statement that all ravens are black is logically equivalent to the statement
that all nonblack things are nonravens. Suppose we hold that generalizations
are confirmed by their instances: the statement that an object a that is A is
also B lends further support to the claim that all A's are B's.35 Examining a
raven and discovering it to be black therefore support the hypothesis that all
ravens are black. However, if we add the plausible claim that a statement e
that confirms a statement h confirms any h' logically equivalent to h, then
we reach apparently counterintuitive conclusions. Suppose that we come to
accept the statement that a certain object that is a shoe is white. By the
principle about instances, this confirms the hypothesis that all nonblack things
are nonravens. Now, from the principle about confirmation of logical equiv-
alents, we can conclude that it supports the hypothesis that all ravens are
black. But surely that is dubious. Observing white shoes is not a cheap way
of increasing reasonable confidence in the generalization about ravens.36
Viewing induction as eliminative enables us to solve the puzzle. The prob-
lematic idea is that any positive instance of a generalization lends it support.
From the eliminativist perspective, instances support generalizations by elim-
inating potential rivals. An instance that does nothing to eliminate a potential
rival does nothing for the generalization. No report of the whiteness of a shoe
does anything to eliminate rivals of "All ravens are black"; some (but not
all) reports of the blackness of ravens eliminate rivals and thus lend support
to the generalization.
Consider the puzzle in more detail. Suppose that the background condi-
tions derived from the explanatory schemata of practice are C1, . . . , Ck, that
35. Here I avoid specifying the order in which the properties of a are recognized. Some
commentators on the raven paradox (e.g., Mackie 1963) have held this to be especially
significant.
36. The paradox has spawned an enormous secondary literature, containing a large
number of ingenious ideas. For a review of some prior attempts at solution, see the dis-
cussion in (Horwich 1982). There are some affinities between my own approach and that
adopted by Horwich, which ultimately derives from an insight of Janina Hosiasson-
Lindenbaum (1940).
The Experimental Philosophy 243
prior practice yields conclusions about the absence of interactions that permit
them to be treated separately, and that the instances already sampled reveal
that ravens meeting conditions Cm+1, . . . , Ck are black. The generalization
under focus is
where D is one of the Ci.37 Some of these rivals have already been eliminated.
Those that remain are of the preceding form with D as one of C1, . . . , Cm.
To obtain further support for the generalization we need reports of objects
that satisfy R, B, D, where D is one of C1, . . . , Cm. White shoes and lots of
black ravens don't do the trick.
The original puzzle was generated by using the logical equivalence prin-
ciple, and, specifically, by considering the hypothesis
where D is among the Ci, then they are logical equivalents of our original set
of rivals, and the exercise in which we are engaged is one of finding incon-
venient logical equivalents for the statements under focus in the last para-
graph. A report of an object can eliminate one of the rivals (while remaining
compatible with the generalization) provided that that object satisfies D, B,
R. This, of course, is our old condition, and, since white shoes do not measure
up, they confer no support. Providing white shoes with confirmatory power
requires consideration of a different set of rivals. Thus, if prior practice
generated background conditions E1, . . . , En so that there were rival hy-
potheses of the forms
where F is among the Ej, then a report of an object would eliminate a rival
(while remaining compatible with the generalization) if that object satisfied
- B, R, F. In principle it is possible for a white shoe to accord with this
demand. We can draw a conditional conclusion: if prior practice generates
rivals of the most recent form, then it is possible for observation of nonblack
nonravens (maybe even white shoes) to lend support to the generalization.
However, for prior practice to generate such rivals, there would have to
37. The rivals are contrary generalizations about the color of ravens. They can take
the form presented in the text or the more specific form of specifying the colors of variously
characterized subsets of the ravens. The argument of the text goes through on the second
interpretation as with the first.
244 The Advancement of Science
and, because the sample we have examined includes only objects that are
- U, we have managed only a pseudoelimination. The remedy would be to
discover an object that is R, B, U. Of course, since we have no idea what U
is, we cannot do this systematically. We simply amass objects that are R, B,
in the hope that if there is an unknown relevant property it will show up.
Finding white shoes can play no part in this process.
Let me close this section by addressing a different question. Methodol-
246 The Advancement of Science
ogists sometimes debate the relative virtues of prediction and accommodation.
According to one tradition, hypotheses only gain support from the novel
predictions they make; their ability to accommodate statements that were
already accepted is irrelevant to their status. The rival view contends that
prediction and accommodation should be equally valuable; the order in which
information is acquired is epistemically irrelevant. Which of these is right?
Those who take a dim view of accommodation suggest that it is too easy
to make up a hypothesis that will fit antecedently known statements. Because
this game is so straightforward, no credit can accrue to those who achieve
success at it. By contrast, presenting a hypothesis that is then able to predict
something unexpected is a risky business, and victory here redounds to the
credit of the hypothesis.
Opposing these ideas is the contention that evidence is evidence, irre-
spective of the order in which it is acquired. Why should it make a difference
to the credentials of h if its consequences e1, . . . , em were all known in advance
or if some of themsay ek+1, . . . , emwere first discovered after h was
proposed? (Practitioners of historical sciencesfor example, evolutionary
biologistsare likely to be sympathetic to this question.) After all, in both
instances, the total set of statements available for assessing h at the end of
the evaluation period is the same.
From the eliminativist perspective I have been developing, we can sustain
this positive view of accommodation while explaining why the worries about
accommodation might sometimes be well founded. Suppose that prior practice
generates a set of rivals to h, {h1, . . . , hn}. The critical issue is whether, in
combination, the statements e1, . . . , em eliminate all the hi. If they do, then
in neither case is there any basis for doubt about h other than the residual
doubts about the completeness of prior practice. If they do not, then in neither
case should h be accepted, for there is an outstanding uneliminated rival.
Surprising novel predictions may be striking, but the touchstone of their
evidential force is the ability to reduce the space of rivals sanctioned by prior
practice. Beyond that, as I have noted, they help to allay residual doubts
about the completeness of the space of recognized rival possibilities.
Accommodating the evidence is often not an easy game, because the
constraints from prior practice are so powerful that they make difficult the
genesis of even one hypothesis that will fit accepted findings (more on this
later). However, when the constraints are lax or when confidence in the
completeness of prior practice is (quite reasonably) low, there is room for
doubt about hypotheses that accommodate accepted results. The problem in
this case is not accommodation itself, but the state of prior practice. I con-
jecture that opponents of accommodation have been moved by examples in
which the constraints from prior practice are lax, and there is no serious
attempt to explore a space of rival hypotheses. That type of reasoning should
be frowned upon, but the troubles should be traced to failure to activate the
eliminative propensity, not to some supposed defect in the strategy of ac-
commodation.38 But, of course, surprising predictions are always welcome
38. The approach indicated here is, I believe, consonant with the views about prediction
The Experimental Philosophy 247
when they can be had, for, in these cases, we discover that something that
might so easily have subverted the constraintsleading to a position in which
there was no available solutionis, in fact, in accord with the constraints.
The result is that our confidence in the legitimacy of the constraints is in-
creased, and, derivatively, residual doubts about the hypothesis (the single
solution) diminish.
6. Underdetermination
For all my pleading on its behalf, the notion that individual scientists can prof-
itably modify their practices by using eliminative induction is likely to seem (at
best) quaint. Surely the idea that there are crucial experiments (or observa-
tions) in science was disposed of long ago, specifically by Pierre Duhem (1906).
Thanks to Quine's revival and development of the Duhemian theme (1951,
1960, 1970), the notion that theories are inevitably underdetermined by expe-
rience has become a philosophical commonplace. Scientists, however, some-
times greet this allegedly mundane point with incredulity. "It's hard enough,"
they complain, "to find one way of accommodating experience, let alone many.
And these supposed ways of modifying the network of beliefs are changes that
no reasonablesane?person would make. There may be a logical point
here, but it has little to do with science."39 I think that the complaining scien-
tists are, at least roughly, right. The Underdetermination thesis, in its usual
guise, is a product of the underrepresentation of scientific practice.
Duhem offered two arguments for the impossibility of using eliminative
induction to justify a hypothesis (Duhem 1906/1953 183-187, 189-190, 211-
212, 216-218). His first claim, less prominent than the Underdetermination
thesis but no less a staple of philosophical orthodoxy, is that any hypothesis
has an infinite set of rivals. My response to this point is already prefigured
in previous discussions. Against the background of prior practice there may
be only a finite set of serious possibilities.40 Consider the extinction of the
dinosaurs. There are any number of logically possible hypotheses that could
account for their demise: perhaps little green aliens exterminated them or
and accommodation advanced by John Worrall (1985). For Worrall and Elie Zahar, what
matters is the evidence that is used in generating the hypothesis, and their concern is that,
in my terms, all the constraints on the hypothesis should come from the evidence. Hence,
I see the general idea that accommodation is unobjectionable when there are severe con-
straints imposed by background practice (and when background practice is, itself, not
problematic) as articulating their insight.
39. Concerns along these lines were voiced by Stephen Jay Gould. In several conver-
sations in the early 1980s, I tried to convince Gould that Duhemian holism spells doom
for any idea of conclusive falsification. He resisted my arguments. I now think that he was
correct.
40. Similar approaches to problems of Underdetermination have been pursued by Lau-
dan and Shapere, both of whom take reliance on background knowledge to be crucial
(Laudan 1984, Shapere 1984). Obviously, much depends on the notion of a "serious pos-
sibility." Much of the rest of this chapter is devoted to explaining how such possibilities
are generated and to illustrating my claims with detailed historical examples.
248 The Advancement of Science
there were spontaneous combustions at all the right places to eliminate all
the females. However, the serious hypotheses are relatively few in number,
and paleontologists spend their efforts in trying to discriminate accounts in
terms of meteoric impact from those that appeal to volcanic eruptions, atmo-
spheric changes (increased oxygen tension), climatic changes, and a few other
factors.41 Why do these deserve greater attention? Because our prior practice
recognizes certain kinds of processes as occurring in nature and not others.
Of course, by abandoning that practice and adopting something different we
could change the assessment of the serious possibilities. There is a potential
modification of practice that would bring into consideration hypotheses that
are presently outside the space of possibilities. As things stand, however,
investigation of the demise of the dinosaurs goes forward under some stringent
constraints.42
Even if the constraints do not manage to reduce the serious rivals to a
finite number, it is still possible for the eliminative strategy to succeed. For
the infinite set of hypotheses may be divisible into a finite number of subsets,
subject to wholesale elimination. (The subsets may, for example, correspond
to ranges of values for continuously varying parameters, and one may be able
to show that the parameters do not fall into some of the ranges. See (Earman
1992) for a convincing illustration). To oppose eliminative induction on Du-
hem's first grounds, it is necessary to suppose that attempts to delineate a
space of rivals encounter systematic failure. This can only be done by em-
phasizing the revisability of the constraints imposed by prior practice. The
Duhemian point must eventually come down to the thesis that we have a
choice of infinitely many rival hypotheses because it is always possible to
introduce newhitherto "nonserious" rivals by giving up parts of our prac-
tice. This, of course, is simply a version of the underdetermination thesis.
Duhem's first challenge to eliminative induction presupposes the success of
the second.
That second challenge is often seen as deadly. Hypotheses, Duhem
claimed, are never tested in isolation (1906/1953 187). Any attempt to confront
a hypothesis with empirical results takes for granted auxiliary hypotheses that
may be revised if things go wrong. So the testing situation is seen as involving
three statements H (the hypothesis taken to be under test), O (a report
accepted on the basis of observation), and A (a conjunction of auxiliary
statements that are required for the derivation of observationally ascertainable
41. To see how the practitioners view the debate, see Scientific American (1990), no.
9 (September).
42. This point about scientific practice is readily accommodated in some accounts of
science. Thus, Kuhn's conception of "normal science" allows for determinate resolution
of issues because of the constraining role of background practice (the "paradigm"). Making
this comparison raises the worry that my critique of underdetermination will be flawed
because of the possibility of breaking free from the constraints in a scientific revolution.
This is a serious challenge, and I shall try to address it later by showing how, even in such
large transitions in the history of science as the chemical revolution of the eighteenth century
and the acceptance of minimal Darwinism in the nineteenth, underdetermination can be
resisted.
The Experimental Philosophy 249
statements from H). Since H,A O, the set {H, A, - O} is inconsistent. Some-
thing has to give. But what is to be amended is, allegedly, not fixed. Scientists
have the options of (i) abandoning H, (ii) abandoning or modifying A, (iii)
refusing to accept - O. Because any of the options can be exercised, there
are any number of systems of beliefs that scientists can square with their total
experiences. In particular, since H can always be retained no matter what H
may be there is an infinite number of total theories each compatible with our
total experience. (To construct the infinite set of theories, simply form a
denumerable infinity of distinct statements and embed each of them in a total
theory.)
The natural response to this point is to contend that not all of this panorama
of possible theories, and not all the responses to experience that underlie
them, are equally good.43 Consider the strategy, explicitly canvassed by Quine
in a famous passage (1951 43), of "pleading hallucination" in the face of
disconfirming observations. Sometimes pleas about the unreliability of ob-
servation are in orderconsiderations drawn from prior practice counsel that
the current observational situation should be viewed with suspicion. But imag-
ine that, in the case at hand, the observational rules of prior practice insist
that the current situation is excellent. If those parts of practice are given no
power to constrain the current options, then the core of the underdetermi-
nation thesis is the following logical truism: any consistent statement can be
embedded in a consistent set of statements. The thesis adds to this the simple
idea that some of the statements in this set express "rules" (carefully chosen)
for evaluating observational situations.
Nobody ought to be astonished by the suggestion that there are numerous
ways of forming sets of beliefs if there are no constraints imposed by obser-
vation, history, or anything else (save logicand, of course, Quine urges the
possibility of revising logic as well [1951 43]). Descartes's program for epis-
temology failed, and the basic point made by the underdetermination thesis
records that failure. However, there is a more interesting version of the
doctrine, at which Quine sometimes hints,44 according to which there will be
alternative ways of revising beliefs even when all the constraints imposed by
methodology are given their due.
To evaluate this doctrine, we first need to formulate it. Let a Duhemian
situation (or predicament) be an occasion on which scientists accept a hy-
pothesis//, accept certain auxiliary assumptions A, where H, A O, and where,
following their normal observational procedures, they would accept the state-
ment - O on the presentation of a stimulus to which they are actually exposed.
I shall suppose that the methodological maxims of accepted scientific practice
43. Again, my approach to this issue is akin to those of Laudan and Shapere. All three
of us recognize grounds for distinguishing systems of belief beyond mere consistency with
"the evidence"although we formulate the distinctions rather differently.
44. The hints are scattered throughout (Quine 1960) in those passages in which he
insists that scientific theories are underdetermined, even when we imagine "an ideal organon
of scientific method." For discussion of Quine's oscillations on the question of whether any
such organon is to be considered in evaluating underdetermination, see (Laudan 1990).
250 The Advancement of Science
might arise from alternative ways of measuring costs of changes recalls the
ideas about shifting standards that have been emphasized by Kuhn and some
of his followers (see Kuhn 1962/1970, 1977 essay 13, Doppelt 1977). I shall
consider the possibilities in turn.
The simple reason for believing that there are many ways of accommo-
dating recalcitrant observations with minimum mutilations is that troublesome
cases can be treated as local exceptions. If I want to escape the {H, A, O}
situation, why cannot I make a small amendment in (OR) to add a defeating
condition that will apply just to the special circumstances that generate the
belief that 0? Or why can I not circumscribe some generalization in A so
that its new range is just like its old except for not including the particular
instance to which it was applied on this occasion? Such solutions by gerry-
mandering only appear cost-free so long as one overlooks the fact that ob-
servational rules and auxiliary generalizations are constrained by the
explanatory dependencies and the views of projedibility present in prior
practice. A claim that there is a local error in investigating things of a very
special type needs to be integrated with our background ideas about the
potential sources of error. That integration cannot succeed if our explanatory
categories provide no way of distinguishing the special type from a much
broader class in which the mode of investigation is to be endorsed as reliable.
Faced with the contention that we simply cannot trust our observation to
record the properties of the shadows of rapidly rotating small circular objects,
the natural riposte is, What's so special about this observational situation?
By contrast, the insistence of some Aristotelians that the telescope could not
reveal the character of the immutable heavens was grounded in distinctions
that were employed in their explanatory schemata.46
Once the constraining character of prior practice is understood, there is
no quick argument for the widespread presence of underdetermination on
Quinean grounds.47 The harder task is to defend against the possibility of
underdetermination that results from rival systems of counting costs. Kuhn
and others have rightly insisted that champions of rival points of view might
assign different significance to different questions. In principle, this can gen-
erate occasions on which the relative merits of strategies for escaping Du-
hemian situations are oppositely assessed because one preserves solutions
valued by one side while the other continues to endorse the solutions valued
by the other. Moreover, I agree that such occasions sometimes arise in the
46. Thus, in the dispute over the veridicality of telescopic observations of the heavens,
the legitimacy of the celestial/terrestrial distinction became a matter for debate. As I noted
earlier, Galileo's defense of the telescope rested in part on undermining the distinction
through appeal to such "naked-eye" phenomena as the novae.
47. In the end, everything will have to turn on close examination of individual cases.
Hence I regard the brief discussion the triumph of Copernicanism in this section, and, even
more, the extended review of the transitions to Darwinism and the "new chemistry" of
Lavoisier, as crucial parts of a philosophical argument. Although each of these examples
has been discussed by many commentators, it seems to me important to combine a look
at the details with the epistemological perspective that I favor.
252 The Advancement of Science
history of the sciences and that debate is sometimes temporarily inconclusive.
The compromise model of the last chapter explicitly allowed for this, while
maintaining that such controversies are ultimately resolved through the pro-
vision by the victor of successes in the terms demanded by the rival.
Faced with claims about rival admissible cost functions, we should look
at the costs on both proposals for emendation and ask whether there is a
coherent conception of scientific significance that underwrites them. For many
writers influenced by Kuhn (for example, Doppelt 1977) it seems that the
recognition of the possibility of counting costs in various ways leaves no
objective grounds for appraisal of rival responses to a Duhemian predicament.
Rash judgments can be tempered by reminding ourselves that there are in-
stances in which preferring one response requires dismissing the significance
of a large number of predictions and explanations and taking debate to revolve
solely around one difficulty. Making that evaluation presupposes a conception
of scientific significance that must itself be defended.
Let me hammer the point home with an analogy. Most of us, at some
stage of our lives, have bought a second-hand (usedor, as the upmarket
advertising has it, "preowned") car. Buying decisions typically involve traips-
ing around various lots and comparing the virtues of a number of possible
vehicles. Few people would be confident that they have ended up making the
best decision. Most would surely recognize that there are competing advan-
tages and disadvantages, and that the final decision is made by weighing some
factors more heavily than others. Rival systems of counting pluses and minuses
might well have generated different decisions, and some of those decisions
might have brought the agents closer to satisfying their preferences. I shall
suppose that those preferences involve wanting a car that will transport people
from place to place, not wanting a car that demands frequent repairs, wanting
a car that will last for a while, and so forth. While acknowledging the existence
of rival schemes of accounting that would yield different sensible decisions,
we should not overlook the fact that some such systems are ludicrous. Faced
with two cars, one with a functioning engine, transmission, good tires, and
the other without engine, transmission, or wheels, we tend to prefer the
former. That tendency would not be overridden for decision makers who were
reasoning well, if the sole virtue of the second car were the fact that its front
right bumper had less rust than the front right bumper of its rival. There is,
of course, a possible scheme of valuation in which presence of rust on the
front right bumper is accorded overwhelming priority. But that scheme cannot
be backed with any coherent conception of significant factors.48
Let us now turn to a problematic example, the debates about Coperni-
canism between 1543 and 1632. As I pointed out in the last chapter, there
48. As the referee insightfully pointed out to me, there is a way of bringing my approach
within the Bayesian framework. Instead of thinking in terms of probabilistic conditional-
ization, one can regard scientists as facing Bayesian decision problems in which the values
of various alternatives are set by just the types of considerations to which I would appeal
in "counting the costs." I shall leave it to Bayesians to decide exactly how much of what
I claim can be accommodated within a natural extension of their views. But see 10.1.
The Experimental Philosophy 253
49. This is one examplealthough the most strikingof the "mathematical harmonies"
on which Copernicus set such store. For others, see (Margolis 1987).
254 The Advancement of Science
51. I would suggest that one of the failures of the Tychonic system to appeal to some
mathematicians and astronomers lay in its inability to furnish any clues to a physical ex-
planation of the motions of the planets. By the turn of the seventeenth century, there are
clear indications that two ideas about celestial influence are gaining currency: large heavenly
bodies have more influence than smaller ones, and bodies have greater influence on closer
neighbors. Versions of these ideas are evident in William Gilbert, in Kepler, and in Galileo.
Against the background of these dynamic hypotheses, the Tychonic system appears hope-
less. For it is a mystery how the Sun can both be subject to the influence of the Earth and
able to influence bodies that are sometimes closer to the Earth than to it. Given both the
vague dynamic ideas, the rotation of the Sun around the Earth could be explained if the
Earth were larger than the Sun, but this would yield trouble for the idea that Mars, Jupiter,
and Saturn (and, on some versions, Venus), which are all sometimes closer to the Earth
than the Sun, rotate around the Sun rather than around the larger Earth. So far as I know,
detailed historical investigation of the role of embryonic dynamical hypotheses in the re-
jection of the Tychonic system has not yet been undertaken.
256 The Advancement of Science
been explored and what possibilities remain open. They make their delib-
erations by relying on parts of prior practice. After all, what else do they
have to go on?
Yet practice can itself be modified. The constraints that favor some lines
of solution and eliminate others can themselves be changed. Our next task
is to investigate how this is possible.
7. Adjusting Constraints
For any individual practice and any set of inconsistent statements that a
proponent of that practice is inclined to accept, we can recognize what I shall
call the escape tree. A path through the tree is determined in the following
way: Start with the set of inconsistent statements. Pick any one of them.
Consider the consequences of deleting this statement from the set and rep-
resent all the losses that would be incurred on the branch. A response to these
losses consists in the addition of further statements that (i) are consistent with
the remaining members of the original set, (ii) compensate for the losses, and
(iii) conform to the constraints derived from practice. A response is threatened
if there is some set of statements that proponents of the practice are inclined
to accept that are inconsistent with the newly added statements. In such cases,
the branch can be continued by attempting to remove the inconsistency from
the new set. If the scientist(s) can think of no way of continuing then the
path is blocked. On the other hand, if at any stage of the branch a response
is unthreatened, then the branch offers an escape from the original predic-
ament and the problem of removing the inconsistency has been solved.
Escape trees presuppose constraints. When do scientists begin to consider
modifying the constraints? Such situations start with an inconsistency whose
escape tree has the following characteristics:
(a) every path through it is blocked.
(b) the scientist believes that it represents the range of possibilities allowed
by the constraints.
(c) the scientist assesses the costs of the least costly path as severe.
52. The weak realism described in 8 of Chapter 5 provides a ready explanation of the
status of the unification principle, but, as admitted there, weak realism has a much more
complex task of reconstructing talk of kinds and causes. Here, as elsewhere in philosophy,
epistemological advantages are purchased at metaphysical cost.
258 The Advancement of Science
checkedfamiliar interfering conditions that have to be eliminated and more
recondite possibilities that cautious scientists considerand the process of
exploration is far more lengthy.
The alternative strategy is to eliminate some claim that does apparent
explanatory or predictive work in other instances and attempt to make good
the losses involved. So, for example, one might try to abandon a hypothesis
H that had been used in providing an explanation of some phenomenon P,
an explanation in accordance with the accepted schemata. Thus the problem
is shifted to explaining P in some way concordant with accepted schemata.
But it may be relatively easy to show that, given beliefs that are invoked in
a vast range of apparently successful explanations, H is the only possibility
for explaining P in accordance with the schemata. Or there may be some
possibility of undermining the argument for the uniqueness of H as a com-
ponent of an explanation of P, by identifying some relatively unemployed
premise in that argument, H'. Abandoning H' would only cost us the expla-
nations of P' and P", and perhaps we could make these up in accordance
with the prevalent explanatory schemata. So the escape tree ramifies further
as the explanation of P' and P" now becomes a problem with increased
significance. Yet exploration of this possibility too may fail, so that the tree
remains blocked.
The elements of scientific decision making are, I contend, quite simple.
A scientist struggles to eradicate inconsistencies, maintain a unified account
of the phenomenaconceived in terms of the background repertoire of con-
cepts and explanatory schemataand project regularities in accordance with
the views about representative samples and projectible predicates embodied
in his practice (views which are, as I have suggested, dependent on ideas
about relevant causal factors and, thus, on the accepted explanatory schemata
of the practice). These elements are far simpler than those typically introduced
in theories of confirmation. I have invoked no intricate logical machinery for
computing degrees of confirmation, no precise assignments of probability to
hypotheses or processes of conditionalization on experiential input.53 Yet, as
should be clear from the discussion of the preceding paragraphs, the activity
of scientific decision making is anything but simple. For the elementary pro-
cesses are combined in elaborate ways as a scientist seeks a path through an
escape tree. A central problem of scientific inference consists in showing how
endeavors to find paths through escape trees yield various types of modifi-
cations of practice.
Consider some of the kinds of changes that may be produced by efforts
to follow a path through an escape tree. First, there may be modifications of
the significance of questions. Suppose, for example, that the attempt to evade
a Duhemian predicament shifts the initial problem to that of finding an al-
ternative explanation for the phenomenon P. This may not only dramatically
increase the significance of the question, Why P?, but also bestow derivative
significance on other questions, answers to which would contribute to the
provision of an explanation of P. So, for example, if a potential explanation
of P is ruled out because it would seem to require values of a particular
quantity that are too high, and if the scientist suspects that the apparatus used
in the past to measure values of the quantity is not as reliable as has been
assumed, then she may come to take as significant the question of how that
quantity can be determined more reliably. Alterations of the perceived sig-
nificance of a question can cascade throughout the appraisal of other problems
and can affect actions taken with respect to instruments and apparatus.
Revised estimates of the significance of a question can also be the direct
product of conceptual reform. The problems scientists set for themselves
depend on their conceptions of the boundaries of kinds. Exploration of the
idea that a single predicate covers two quite different types of entity will
typically lead to the rejection of general questions involving the predicate in
favor of more narrowly specified problems. But how is conceptual reform
itself initiated?
Distinguish three important ways in which scientists may be led either to
introduce new language or to adjust the reference potentials of old terms. (I
do not claim that these are the only such ways or that all others are less
important.) First, one may try to respond to a Duhemian predicament by
finding a generalization about a set of entities. The available information
about these entities may suggest that their instantiation of determinate forms
of a particular determinable property is sufficiently variable and confused to
suggest the possibility that distinctions should be made among them, even
though the scientist does not yet see how to draw such distinctions in terms
allowed by the available explanatory schemata. This kind of conceptual re-
vision may precede modification or further articulation of the explanatory
schemata to underwrite new distinctions, but, in accordance with my earlier
remarks about the loss of unification produced by simply declaring exceptional
cases, it will ultimately prove important to integrate the distinctions with a
set of explanatory schemata. A second style of conceptual reform is generated
by drawing distinctions that are motivated by prevailing explanatory sche-
mata: one distinguishes those A's that are C from those that are not, adducing
some previously unrecognized condition C that exemplifies a type which ac-
cepted schemata view as relevant to the behavior of entities like A. Third,
and perhaps most straightforwardly, conceptual change can be fuelled by
modifications of explanatory schemata. As we change our views about which
260 The Advancement of Science
57. In line with the emphasis on cognitive variation in Chapter 3 and the suggestion
that social interactions among proponents of different individual practices craft cognitive
strategies that are superior to anything that underwrites the beliefs of any single individual
(see Chapter 6), I would see the task of mapping the reasonable ways of articulating escape
trees as typically allowing alternatives. As noted in footnote 55, I defer this project to a
future study, recognizing that the provision of methodological exemplars may ultimately
be all that methodology can offer.
58. This differs only in level of detail from that offered in the notebooks, so, in this
instance, we need not worry whether we are scrutinizing his initial reasoning or a retro-
spective justification.
264 The Advancement of Science
(WA) admits of a strong and a weak reading. On the strong construal, well-
adaptedness conforms to the idea that the Creator arranges for an optimal
distribution of organisms to sites of Creation: among their contemporaries of
the same type, the organisms created at P are the best adapted for living in
P. The weak reading supposes merely that they do well in P, allowing for
the possibility that some of their contemporaries might do even better.
Armed with the strong version of (WA), creationists can tackle (A)-(C).
The treatment of the neighborhood problem begins from the idea that ad-
jacent places are physically similar, and hence call for organisms with similar
traits: the optimally adapted organism in P is likely to be akin to the optimally
adapted organism in P* if P and P* are much alike. Both (B) and (C) can
be handled by recognizing independent creations of the same, or of similar,
species. However, the commitment to the strong form of (WA) introduces
severe problems for the creationist, difficulties that will be highlighted by
(D). Darwin's biogeographical argument begins with an acute diagnosis of
why the strong form of (WA) is too strong.
Chapter XI of the Origin opens with a statement of the anticreationist
thesis: "In considering the distribution of organic beings over the face of the
globe, the first great fact which strikes us is, that neither the similarity nor
the dissimilarity of the inhabitants of various regions can be accounted for
by their climatal and other physical conditions" (346). Darwin follows up
with an array of examples: Old World environments find their counterparts
in the New World, and yet the faunas and floras are quite different; South
Africa, western South America, and Australia have three "utterly dissimilar"
59. Here, of course, we see the force of Darwin's frequent appeals to the goal of unifying
our account of geographical distribution, and of Huxley's desire to reduce the ''fundamental
incomprehensibilities" to the smallest possible number.
266 The Advancement of Science
faunas, despite having extremely similar physical conditions. Much later, in
discussing the distributions on oceanic islands, the attack is focused sharply
as presenting conclusive objections to the strong version of (WA).
In St. Helena there is reason to believe that the naturalised plants and animals
have nearly or quite exterminated many native productions. He who admits
the doctrine of the creation of each separate species, will have to admit, that
a sufficient number of the best adapted plants and animals have not been
created on oceanic islands; for man has unintentionally stocked them from
various sources far more fully and perfectly than has nature. (390)
The displaceable organisms problem spells doom for the strong version
of (WA). Schematically, the creationist is faced with an inconsistent set of
commitments: {G's were created in P, G*'s were created in P*, P and P* are
distinct, G's and G*'s are contemporaries of the same type, organisms created
at a place are better adapted for living there than any of their contemporaries
of the same type, G*'s are better adapted for living in P than are G's}. Faced
with the example of the plants and animals of St. Helena, creationists can
consider lines of escape from this predicament. It is hardly attractive to sup-
pose that the inhabitants of St. Helena were created somewhere else: How
did they get there? What has happened to them at their original place of
creation? Nor is there joy in supposing that the introduced aliens were, in
fact, originally created on St. Helena: What happened to them after their
original creation there? The facts of competition, including near or complete
extermination, make it impossible to deny that the aliens are better adapted
for life on St. Helena than are the natives. Revisionist geography, that would
question the thesis of disjointness, has nothing to recommend it, and it is
similarly impossible to question the fact that the organisms that have competed
on St. Helena are contemporaries of the same type. Under these circum-
stances, the escape tree is blocked in every direction but one. The strong
version of (WA) has to go.
Now, however, we have to consider what happens to the general line of
solution to comparative problems in biogeography. Initially, it seemed that
the creationist would answer questions of the form, Why are G's and not
G*'s created at P?, by instantiating the following schema:
With the observation that the similarities among neighboring organisms out-
run the similarities of their environments, Darwin exposes the inadequacy of
creationist solutions that trace similarities of form to similarities of environ-
mental demand.
But elimination of the creationist program will only work as an argument
for Darwin's own solution if he can show that there are no promising available
alternatives and that his own proposal is not subject to kindred severe criti-
cisms. We have already examined the case for the first point. For any given
species, the options of descent from a prior species or creation de novo appear
to be exhaustive. Hence, we must choose between some version of descent
with modification and some form of creationism. As I have already suggested,
minimal creationism is untenable, except possibly as a position of last resort,
a confession of ignorance in reaction to the failure of all positive efforts to
explain the distributions of plants and animals. So the creationist must find
some substitute for the idea of creation in response to environmental demand
and the strong principle (WA) to which it gives rise. Lacking any clues about
how to develop the needed account, it is hard to retain creationism as a live
possibilityunless, of course, its rival is in trouble, and we find ourselves in
268 The Advancement of Science
the dismal situation of having eliminated all the initially promising candidates.
The crucial issue, then, is the viability of Darwin's own account, and the bulk
of chapters XI and XII is devoted to responding to problems and inconsistency
predicaments. He faces apparently serious difficulties generated by problems
(B) and (C).
We can begin to represent the inconsistency predicament generated by
the barrier problem, (B), by noting that Darwin is committed to instances of
statements of the following forms:
[PA] If R and R* are presently mutually inaccessible for G's and G*'s
then it would have been impossible for the common ancestor of G's and
G*'s, GO, to radiate into both R and R*.
60. It is worth noting that Darwin is here pursuing both problems (B) and (C) by
effectively supposing that barrier problems can (often? always?) be tackled by recognizing
a continuous range of the ancestral species on both sides of the region in which a barrier
emerges. His remarks also contrast his own line of solution with creationist approaches to
(C).
The Experimental Philosophy 269
Investigation of the hard cases will require Darwin to look both at the pos-
sibilities of geographical change and at the dispersal powers of organisms.
On the former score, he canvasses various ways in which islands may become
linked to continents, bodies of water joined, changes of climate have allowed
a "high road for migration" (356). Yet, contrary to the enthusiasm of some
of his contemporaries, Darwin demands that hypotheses about past changes
in geographical position must be firmly founded in hypotheses about presently
acting geological processes (357).
The second part of Darwin's strategy, to which he devotes greater length,
explores the ways in which current organisms can be transported. Focusing
on the apparently difficult example of the dispersal of plants across sea water,
he reports a series of experiments to test the ability of seeds to germinate
after floating on sea water for various periods. The argument deserves quoting
at some length.
It is well known what a difference there is in the buoyancy of green and
seasoned timber; and it occurred to me that floods might wash down plants
or branches, and that these might be dried on the banks, and then by a fresh
rise in the stream be washed into the sea. Hence I was led to dry stems and
branches of 94 plants with ripe fruit, and to place them on sea water. The
majority sank quickly, but some which whilst green floated for a very short
time, when dried floated much longer; for instance, ripe hazel-nuts sank
immediately, but when dried, they floated for 90 days and afterwards when
planted they germinated; an asparagus plant with ripe berries floated for 23
days, when dried it floated for 85 days, and the seeds afterwards germinated:
the ripe seeds of Heliosciadium sank in two days, when dried they floated
for above 90 days and afterwards germinated. Altogether out of the 94 dried
plants, 18 floated for above 28 days, and some of the 18 floated for a much
longer period. So that as 64/87 seeds germinated after an immersion of 28
days; and as 18/94 plants with ripe fruit (but not all the same species as in
the foregoing experiment) floated, after being dried, for above 28 days, as
far as we can infer anything from these scanty facts, we may conclude that
the seeds of 14/100 plants of any country might be floated by sea-currents
during 28 days and would retain their power of germination. In Johnston's
Physical Atlas, the average rate of the several Atlantic currents is 33 miles
per diem (some currents running at the rate of 60 miles per diem); on this
average, the seeds of 14/100 plants belonging to one country might be floated
across 924 miles of sea to another country; and when stranded, if blown to
a favorable spot by an inland gale, they would germinate. (359-360)
Darwin apologizes for his "scanty facts," and indeed, if his task were to
establish precise rates of dispersal of plant species, the trials he performed
are obviously too crude. But the goal is the far more modest one of warding
off the threatened inconsistency obtained by instantiating the schematic sen-
tences presented earlier by taking the G's as species of plants and R and R*
as continent and oceanic island. To block the inconsistency it is enough to
show that one of the contributing claims can be challenged, and Darwin's
explorations permit him to cast doubts on the efficacy of the alleged barrier.
This is only the opening wedge of the argument. Sensitive to the point
270 The Advancement of Science
that the means of transport he has canvassed may be too tenuousor that
they may not allow for transport of the right speciesDarwin indicates a
number of other ways in which the threatened inconsistency could be re-
moved. Stones embedded in the roots of trees may enclose portions of seed-
bearing dirt and offer complete protection against being washed away; drift
timber can thus serve as a vehicle for seeds. Birds can swallow seeds, which
are excreted whole and which will germinate. Birds also carry dirt particles
on their beaks or their claws. Icebergs "are known to be sometimes loaded
with earth and stones" (363). When we recognize the variety of ways in which
seeds can be dispersed, Darwin concludes, we should think it to "be a mar-
vellous fact if many plants had not thus become widely transported" (364).
As I read this section of the Origin, Darwin has responded to a threat of
inconsistency by sketching an escape tree with several open branches. He has
not demonstrated that each of the problematic instances can be resolved. But
he has provided grounds for thinking that the problems may not be insuper-
able, that there are resources within his recommended practice to find relief
from inconsistency. By assembling his catalogue of potential modes of trans-
port, Darwin indicates the lines along which solutions can be found and so
provides corrigible grounds for thinking that solutions are available. Given
his successes with other problems (for example with (A)), the existence of
apparent barriers offers no block to the acceptance of his proposal for bio-
geographical explanation.
I shall close by considering Darwin's response to (C), the disconnected
ranges problem, where his discussion focuses on a very particular example
and thus supplements the general outlining of possible solutions with a con-
crete achievement. He begins by suggesting that the case he will treat poses
an exceptionally severe challenge to his approach to biogeography.
The identity of many plants and animals, on mountain-summits, separated
from each other by hundred of miles of lowlands, where the Alpine species
could not possibly exist, is one of the most striking cases known of the same
species living at distant points, without the apparent possibility of their having
migrated from one to the other. (365)
Darwin's predicament can be identified by the following set of schematic
sentences:
When we take the G's to be alpine species of plants and animals, there are
numerous instantiations of these schematic sentences which commit Darwin
The Experimental Philosophy 271
to the first three claims. If he is to escape inconsistency, he must therefore
challenge the relevant instances of [PA*].
This is carried out in the section "Dispersal during the Glacial Period"
by adducing geological evidence for large changes in the earth's climate.
The ruins of a house burnt by fire do not tell their tale more plainly, than
do the mountains of Scotland and Wales, with their scored flanks, polished
surfaces, and perched boulders, of the icy streams with which their valleys
were lately filled. So greatly has the climate of Europe changed, that in
Northern Italy, gigantic moraines, left by old glaciers, are now clothed by
the vine and maize. Throughout a large part of the United States, erratic
boulders and rocks scored by drifted icebergs and coast-ice, plainly reveal a
former cold period. (366)
During this previous cold period, Darwin claims, arctic species of plants and
animals would have been able to extend their ranges throughout much of
Europe and North America. But, as the climate grew warmer again, they
would have been displaced from the new additions to their ranges by the
return of temperate organisms, previously driven south by the advancing cold.
As the warmth returned, the arctic forms would retreat northward, closely
followed up in their retreat by the productions of the more temperate regions.
And as the snow melted from the bases of the mountains, the arctic forms
would seize on the cleared and thawed ground, always ascending higher and
higher, as the warmth increased, whilst their brethren were pursuing their
northward journey. Hence, when the warmth had fully returned, the same
arctic species, which had lately lived in a body together on the lowlands of
the Old and New Worlds, would be left isolated on distant mountain-summits
(having been exterminated on all lesser heights) and in the arctic regions of
both hemispheres. (367)
So Darwin provides evidence against [PA*], while simultaneously giving an
explanation of the apparently anomalous distribution.
Darwin follows up this basic story with some detailed discussions of par-
ticular instances of distribution, designed to show that the glacial dispersal
story not only resolves the initial difficulty but also accounts for particularities
of distribution that have appeared peculiar on extant accounts of biogeography
(see, for example, 376-379 on the distribution of temperate organisms in
mountainous regions of the tropics). Even without probing his further ar-
guments, I think that enough has been said to show the prevalence of those
modes of reasoning discussed in previous sections. Darwin's overall strategy
is to proceed by eliminating rival hypotheses. In doing so, he formulates
inconsistency predicaments for his opponents and shows how the escape trees
from those predicaments are blocked. He also responds to the inconsistency
predicaments that threaten his own account, either showing in some detail
how there is a line of escape (the discussion of the alpine fauna and flora) or
at least indicating the possibilities of amending the inconsistent set without
epistemic loss (the treatment of methods of dispersal). Notice finally that
there is no serious threat of underdetermination. Although there are logically
272 The Advancement of Science
possible rival hypotheses at any number of points in Darwin's argument, none
of them has any plausibility, given the state of practice in natural history from
which Darwin begins. Of course, one could suppose that the cases in which
new arrivals have displaced native inhabitants involve some hushed-up con-
spiracy on the part of colonizing humans, or that the Welsh mountains were
scored by the hand of God (or the devil), or that birds used to secrete digestive
juices that would destroy any seeds they might swallow. One could suppose
that an evil demon has contrived the distribution of plants and animals to
deceive us. Part of Darwin's achievement is to set forth an argument that
leaves extravagant hypotheses akin to extreme forms of skepticism as the
only refuge for his opponents.
61. Notable representatives of this tradition are Conant (1957), Partington (1965), and
Partington and McKie (1937-38).
The Experimental Philosophy 273
that the framework offered in earlier chapters provides a way of focusing this
vexed issue. The work of Lavoisier and his associates certainly brought about
major changes in a number of components of chemical practice: the language
of chemistry was reformed, questions about the composition of substances
and the weight relations among constituents were given far greater significance
than hitherto, new schemata for answering these questions were adopted,
instruments and apparatus for performing chemical experiments were enor-
mously refined, and Lavoisier's famous "principle of the balance" furnished
a new standard for assessing experiments. Appreciating the magnitude of
these changes, we can readily understand why some of the protagonists saw
themselves as living through a revolutionary period in chemistry (although,
for obvious historical reasons, the vocabulary of "revolution" was more likely
to occur to them than to workers in other periods). In line with my approach
in earlier chapters, I suggest that nothing much hangs on whether we apply
or withhold the label.
Part of the story does involve the repudiation of accounts of combustion,
reduction, and other chemical reactions based on appeals to phlogiston. So,
while we cannot identify the changes wrought by Lavoisier with the overthrow
of the phlogiston theory, one facet of his work consists in replacing phlogis-
tonian explanations with different analyses. Mindful of the motivations for
the compromise model, adduced in the last chapter, we must be careful to
allow that the availability of superior cognitive processes supporting Lavoisier
and opposing the phlogistonians may vary with time. Moreover, we do not
have to assume that repudiation of phlogistonian analyses necessarily involved
acceptance of Lavoisier's accounts of all reactions: there are some parts of
Lavoisier's chemistry about which he and his colleagues admit the tentative
character of their conclusions. The thesis that I shall try to illustrate is that,
by the early 1790s, Lavoisier and his co-workers had made available cogni-
tively superior forms of reasoning supporting the thesis that there was no
coherent phlogistonian analysis of central chemical reactions and that there
was a coherent rival conception of those reactions based on the entities (oxy-
gen, hydrogen, and so forth) which the French chemists recognized.
Both the opposing stories of the chemical revolution have a clear virtue
and a clear defect. Conant enables us to understand why there was a pro-
nounced trend to Lavoisier in the 1780s, so that, by the mid-1790s, virtually
every chemist had abandoned the phlogiston theory.62 However, the expla-
nation of easy consensus is purchased at the price of making those who resisted
Lavoisier at all seem unintelligent. By contrast, Kuhn, and, following him,
Doppelt, are determined to understand how extremely astute people could
62. As is well known, Priestley remained unconvinced until his death in 1804. The
parade of "converts" to Lavoisier is, however, impressive: in 1785, Berthollet; in 1786,
Fourcroy; in 1787, Guyton de Morveau; in 1791, Richard Kirwan. See (Perrin 1988b) and,
for the situation in Germany, (Partington and McKie 1937-1938). Priestley's refusal to join
the band should be put into context by recognizing the breadth of his intellectual interests
and the important changes that took place in his personal life in his last two decades. See
(Schofield 1967).
274 The Advancement of Science
engage in debate over a period of two decades and still not reach consesnsus.
Their answer is that the available evidence could not force a decision, but
this answer carries with it an inability to understand the overwhelming con-
sensus of the 1790s. Doppelt, who is more explicit than Kuhn, suggests that
decisions are ultimately made between "equally reasonable" points of view,
because of "various sociological, psychological, biographical, and historical
factors" (1978 54), but he does not identify which such factors were operative
in the consensual repudiation of phlogiston.
According to Doppelt, the chemical revolution serves as an outstanding
illustration of Kuhn's epistemological relativism.63 The inability of reason and
evidence to resolve major scientific disputes is supposed to rest on shifts in
standards, whereby scientists make alternative, equally reasonable, decisions
about what counts as science, what problems are important to solve and what
counts as a solution, as well as on losses of data, instances in which experi-
mental or observational results that were accounted for in earlier discussions
are simply discarded because they cannot be explained from the new point
of view. So, in Doppelt's story, the phlogiston theory
explained the common properties of the metals as due to their possession of
phlogiston, lacking in their ores. . . . In effect, the new "quantitative" chem-
istry of Lavoisier and Dalton abandoned any concern for these questions and
these observational datawhose treatment constituted the main achievement
of the earlier model of chemistry. (1978 43)
He goes on to suggest that the problem of understanding weight relations
assumed far greater importance within Lavoisier's new chemistry than within
the phlogiston theory, and that, in consequence, although phlogistonians
could recognize the anomaly that metallic calxes (supposedly resulting from
metals through the loss of phlogiston) weigh more than the metals, they could
downplay this problem as insignificant (1978 44).64
The story of what occurred among chemists between 1770 and 1790 is far
richer and more interesting than either the Conant or the Kuhn/Doppelt
accounts recognize. It is, I believe, possible to honor both desiderata, showing
how highly intelligent people could have disagreed for so long while under-
standing how the experimental evidence and the construction of reasoning
based on that evidence could ultimately decide the credentials of phlogiston.
63. Of course, Kuhn has repeatedly denied that he espouses relativism, and it is there-
fore unclear whether or not he would endorse the arguments that Doppelt ascribes to him.
Nevertheless, as Doppelt's extensive quotes show (1978 40, 48), it is hardly inexplicable
that Kuhn's (1962/1970) has been read as a defense of relativism. I shall leave it to others
to decide whether Kuhn is committed to the ideas that Doppelt derives from him. My aim
is to show that those ideas make little contact with the reasonings of the chemists of the
late eighteenth century.
64. As John McEvoy notes (1988 199), Doppelt's exposition presents a philosophical
thesis about the history of science with remarkably little historical evidence. As I shall
argue shortly, a closer look at Doppelt's primary example shows how far his account is
from the phenomena it is supposed to analyze.
The Experimental Philosophy 275
65. Here I oversimplify. Lavoisier, Priestley, and their contemporaries wanted to give
chemical analyses of other reactions, and to integrate their analyses with explanations of
vital processes. For Lavoisier, see Holmes (1985) and the middle part of Lavoisier (1789).
For Priestley, see his (1775).
66. I shall return to the issue of Lavoisier's own difficulties in articulating his "anti-
phlogistic system" at the end of this section. My account will be heavily dependent on
Holmes's pioneering excavations of Lavoisier's laboratory notebooks.
276 The Advancement of Science
67. For two examples among many, see Kirwan's discussion of reactions with nitrous
acid (Kirwan 1782/1789 76-115) and (Priestley 1775 vol. III 82). McEvoy (1988 203-204)
makes similar points about the thesis (defended in Schofield 1967) that Priestley had a
methodological difference with Lavoisier about the use of the principle of the balance sheet,
and he argues cogently that theses about Kuhnian incommensurability cannot be illustrated
by reference to this example.
The Experimental Philosophy 277
iliary assumptions is typically not free, and, in the case at hand, both physics
and chemistry provide constraints that close off the escape trees generated
by the notion of the levity of phlogiston. Lavoisier's (1783) points out that
there are experiments which demand positive weight for phlogiston (more
exactly, he shows that phlogistonian accounts of some experiments are com-
mitted to the ponderability of phlogiston). Moreover, efforts to integrate the
levity of phlogiston with gravitational theory broke down in accounting for
phenomena of falling bodies and pendulum motions.68 The difficulties for
those who endeavored to satisfy the principle of the balance without conceding
that something is absorbed in calcination were increased by Lavoisier's ex-
periments showing the proportionality of the weight gain to the volume of
air lost. Thus it is hardly surprising that, for many phlogistonians, including
the most prominent and influential champions of the "old doctrine," the idea
of phlogiston with negative weight has no appeal whatsoever: as early as 1775,
Priestley announces that he "never had any faith at all in that doctrine of the
principle of levity" (1775 vol. II, 312).
Armed with these points, we can begin to analyze the debate over the
credentials of phlogiston. In the first phase, which occurs in the 1770s, La-
voisier shows that there is a weight gain in the calcination of metals, and he
demonstrates that the weight gain equals the weight loss of the air.69 At this
point, phlogistonians do not announce that they have qualitative explanations
that Lavoisier cannot emulate, they do not question his principle of the "bal-
ance sheet," they do not (with few exceptions, chemists who quickly find
themselves in a thicket of difficulties) campaign for the levity of phlogiston.
Instead, they do something that is far more reasonable: to wit, accept La-
voisier's claim that something from the air is absorbed and try to combine
this concession with the traditional idea that phlogiston is emitted.
I shall consider the structure of this transition in phlogistonian accounts
shortly. Before doing so, I want to note that, at this stage of the debate, in
the mid-1770s, Lavoisier has not yet arrived at a coherent account of what
occurs in the various reactions that he and his interlocutors have investigated:
calcination, reduction of calxes, acid-metal reactions. There is room for gen-
uine disagreement at this point of the controversy because nobody knows how
to offer a consistent treatment of all the reactions on which the parties have
data. In particular, nobody knows whether or not phlogiston emission can be
combined with absorption of a constituent of the air to yield a general analysis
68. See the discussion of Gren's attempt to assign negative weight to phlogiston and
Mayer's critique of it in (Partington and McKie 1937-1938 II 33-38).
69. Thus the 1774 memoir "On the Calcination of Tin in Closed Vessels" proceeds by
showing first that the weight of the whole system remains constant through the experiment,
second that the weight of the excess air introduced to fill the vessel after calcination has
occurred is equal to the weight gain of the calx. Lavoisier writes: "Thus, in this operation,
more air is found in the vessel than before the calcination, and it is evidently this excess
of air that yields the gain in weight; if therefore this same increase of weight is found in
the metal, it will be shown [prouvee] that the excess of air which came in has served to
replace the portion that was combined with the metal during calcination, and which in-
creased its weight:" (1774 112). Lavoisier goes on to report the equality.
278 The Advancement of Science
of combustion (and whether this can be extended to treatments of reduction
and acid-metal reactions).
By the mid-1780s, however, Lavoisier has fashioned a general account
which deals, in a unified and consistent way, with a far greater range of the
experimental results than any extant version of the phlogiston theory. More-
over, he is able to expose systematic problems with the particular phlogis-
tonian accounts developed by Priestley, Cavendish, and Kirwan. This is not
to say that his own analysis is free of problems. Notoriously, Lavoisier had
some trouble in integrating his treatment of chemical composition with his
views about heat, and his discussions of some acids were, as he conceded,
tentative.70 There was still work to be done, after 1790, in showing that
Lavoisier's preferred schemata could be instantiated with respect to some
phenomena, but, well before this date, he and his colleagues had shown that
they could apply their schemata to a class of instances well beyond the range
of the rival phlogistonian approach.
Let us now look more closely at the kinds of reasoning involved both in
the resolution of the initial phase of the controversy in the 1770s and in the
abandonment of phlogiston in the late 1780s. We begin from the experiments
on calcination in sealed vessels, which seem to show the following:
[1] The sum of the weights of the reactants equals the sum of the weights
of the products.
[2] Weight of calx is greater than weight of metal.
[3] The air in the vessel decreases in volume during the reaction, and the
weight loss of the air equals the weight gain of the calx.
Phlogistonians will find these conclusions troubling, for they seem to support
the idea that combustion is a process of absorption, not one of emission. How
can that consequence be avoided?
One suggestion might be that the inference from [1] to the closure of the
system is faulty. But this is hard to sustain. If the proposal is that ponderable
70. I should note that, while Lavoisier's official doctrine is that heat is a subtle fluid,
caloric, that can insinuate itself among the particles of matter (thereby causing transitions
from the liquid to the gaseous state), there are passages in which he expresses a more
ecumenical view. In a remarkable passage in the Traite of 1789 he suggests that caloric
doesn't have to be regarded as a real substance: "It suffices, as will be better understood
on reading what follows, that it should be some kind of repulsive cause that scatters the
moelcules of matter, and one can thus regard the effects in an abstract and mathematical
fashion" (1789 19). We should thus beware of assuming that Lavoisier's treatments of
chemical reactions encountered problems caused by his commitment to the caloric theory
of heat.
Similarly, although Lavoisier claims that oxygen is found in all acids, he explicitly bases
this on experiments on carbonic, sulfuric, and phosphoric acids. "We have not yet succeeded
in decomposing and recomposing all the acids; but we are assured at least that oxygen is
a principle common and necessary to the formation of all those whose composition we are
acquainted with. . . . But when from these particular facts [the analyses of carbonic, phos-
phoric, and sulfuric acids], the general induction is made that oxygen is a principle common
to all acids, the consequence is founded on analogy, and here it is that the theory com-
mences" (Lavoisier's note in Kirwan 1789, 19-20).
The Experimental Philosophy 279
matter escapes from or comes into the system, then it must be explained how
this transaction always occurs so as to give the appearance of equality within.
On the other hand, if the suggestion is that imponderable matter enters or
exits, then it is far from clear how this will solve the problem of the observed
weight relations.
Asserting that phlogiston has negative weight allows the phlogistonian to
accommodate [2], but this assertion now has to be brought into conformity
with the phenomena of mechanics. Moreover, given that Lavoisier is able to
show first that the weight gain of the calx varies with the diminution of the
volume of the air in the vessel, and then that the weight gain of the calx is
equal to the weight loss of the air, resistance to the idea that there has been
an absorption has to account for the loss in weight of the air and the con-
comitant loss in volume. If it is proposed that the weight loss of the air results
from the fact that the air has absorbed phlogiston (with negative weight) then
the diminution of volume must be accounted for by hypothesizing that the
absorption of phlogiston by the air produces a gas which occupies a smaller
volume. Now phlogistonians must explain why the diminution of volume is
such that the weight of that volume of air agrees with the weight gain of the
calx. The hypothesis that phlogiston has negative weight thus faces difficulties
caused by constraints from physics and simultaneously supposes a mysterious
coincidence of values.71
Thus, given the background assumptions shared by members of the phlo-
giston theory (and by their opponents), assumptions that cannot be replaced
without serious epistemic losses, the conclusion to be drawn from [l]-[3] is
[4] When metals are calcinated in air, something is absorbed from the
air.
71. These difficulties are concisely noted by Lavoisier in his retrospective account of
the analytic situation in 1772, when he views himself as having established the absorption
thesis (my [4]). Writing in 1792, he comments on Guyton de Morveau's proposal that
phlogiston has "less weight than the atmospheric air" so that there would be the appearance
of weight gain through loss of phlogiston when the substances were weighed in air.
This explanation would have been tenable, if the increase in weight of the metallic
oxides had not been equal to that of the air displaced, or, which comes to the
same thing, if it had disappeared when they were weighed in vacua. But this
increase is much too large to attribute to that cause, since it amounts, in some
metals, to about one third of their weight. One must thus either abandon the
explanation given by Guyton de Morveau, or go so far as to suppose that phlogiston
has negative weight, a tendency to distance itself from the centre of the earth, a
supposition that will be found in contradiction with all the facts maintained and
recognized by the disciples of Stahl. (1792 102)
Although all the main points of the defense of (4) are assembled here, Lavoisier's
reasoning seems to make an interesting conflation: the point about the weight of the
displaced air has broader scope than the weighing experiment he suggests for discrediting
Guyton de Morveau's proposal and can be deployed to attack the doctrine of negative
weight (a doctrine that Lavoisier dismisses by pointing to its peculiar physical
consequences).
280 The Advancement of Science
However, it is possible to combine [4] with the fundamental principle of the
phlogiston theory, namely
This idea can be developed in one of two ways. Either the phlogiston is lost
to the residue of the air used in the calcination, or the escaping phlogiston
combines with the entering constituent of the air, which, joining to the de-
phlogisticated metal, produces the calx. We can represent these articulations
as follows:
On the basis of the line of reasoning we have rehearsed so far, the principal
versions of the phlogiston theory after the late 1770s accept [6] and, in con-
sequence, either [6a] or [6b]. The debate of the 1780s is thus between phlo-
gistonians who espouse some version of [6a] or [6b] and antiphlogistonians
(initially Lavoisier, later Lavoisier and other French chemists) who try to
show that neither [6a] nor [6b] is tenable.
Quite plainly, the experiments to which Lavoisier appealed in supporting
[l]-[3] do not suffice to discredit versions of [6]: [6] is explicitly designed to
accommodate them. By 1777, Lavoisier had arrived at an alternative story,
one familiar to us. The air, he suggested, consists of two parts. One, "emi-
nently respirable air" (or, sometimes, "vital air") is absorbed by metals during
calcination; Mr. Priestley, Lavoisier notes, has "very improperly called [this]
dephlogisticated air" (1777 184). The other part, "mephitic air" is a "mo-
phette," which "contributes nothing to phenomena of combustion" (1777
192). Lavoisier indicates how this approach will explain the established results
of calcination, and he announces that he is engaged on a series of experiments
that will oppose the phlogistonian accounts of Stahl and of Priestley (1777
190). Six years later, at the beginning of Reflections on Phlogiston, Lavoisier
alludes to the state of his argument in 1777, by viewing it as dependent on
the use of Occam's razor. As of 1777, he suggests, he could show that the
phenomena of combustion and other types of calcination could be understood
without invoking phlogiston, so that this entity was purely hypothetical, un-
necessary, and, according to the principles of "good logic," should be dis-
The Experimental Philosophy 281
missed (1783 623). But the new attack is to go further and to cut deeper.
Lavoisier tries to show that, in accounting for the wealth of experimental
results that are now available, defenders of phlogiston become enmeshed in
contradictions.
As we might expect, Lavoisier's assessment of the epistemic state in chem-
istry in the late 1770s reflects his partisan perspective, but, I shall suggest, he
is correct in recognizing that there is an important shift in the 1780s. Having
established [4] in the way that we have considered, Lavoisier could offer, in
1777, a general explanation of processes of calcination in terms of absorption
of part of the atmosphere. But this did not eliminate the possibility of explaining
those processes by adopting [5] and [6]. Moreover, the apparent simplicity of
Lavoisier's account of combustion might readily be offset by difficulties with
other experimental findings. Calcination was not the only process of interest
to or recognized by the late eighteenth century chemists. Hence, the Occamite
suggestion to dismiss phlogiston could easily seem premature, when the ac-
counting for so few chemical reactions was in hand.72
The task facing both phlogistonians and antiphlogistonians in the late 1770s
was to provide analyses of a variety of chemical reactions, including calcin-
ation, reduction of calxes, and the acid-metal reactions. Both sides were
committed to the following schema:
Chemical Composition
[C1] In reaction R the reactants are R1 . . . Rm in relative proportions
by weight r1, . . ., rm.
[C2] The constituents of Ri are Si1,. . . , Siki with relative proportions
by weightSi1,. . . , Siki.
[C3] Under the temperature conditions of reaction R, the affinities of
the constituent substances S11, . . . , Smkm for one another are. . . .
Therefore
[C4] The products of R have the chemical constituents Tj1, . . . , Tjkj with
relative proportions by weight tj1, . . . , tjkj, and the products are in the
relative proportions p1, . . . , pn.
[C5] Pj is the substance which has the constituents Tj1, . . . , Tjkj with
relative proportions by weight tj1, . . ., tjki.
Therefore
[C6] The products of the reaction are P1, . . . , Pn in relative proportions
by weight p1, . . . , pn.
72. As we shall discover, knowledge of the apparent reactants and products of other
chemical reactions could make phlogistonian analyses seem attractive. Enormous work
would be needed to lessen these attractionsand, as we follow the twists, turns, and
complexities of the debate, it will hardly seem surprising that resolution of the controversy
took so long.
282 The Advancement of Science
In this schema, all parties to the dispute agree to the principle of the balance
sheetin other words, that the amounts of the constituent substances must
agree in both reactants and products. Phlogistonians maintain that all assign-
ments must be consistent with [5] and [6]. Antiphlogistonians contend that
these assignments must be compatible with
73. Here I ignore complications stemming from Lavoisier's ideas about the combinations
of bases of substances with caloric. On his official view of heat, vital air consists in the
union of the "basis" of vital air with caloric. Not only does the simplification not affect
the epistemological issues concerning the dispute between Lavoisier and the phlogistonians,
but, as mentioned in note 70, Lavoisier himself saw that the assumption of heat as a
substance could be omitted from his system.
The Experimental Philosophy 283
The versions of [8] most prominent in the 1780s and 1790s are those canvassed
by Kirwan in 1782: Cavendish supposes that Z = water; Kirwan himself
believes that Z is often "fixed air" (carbon dioxide) but sometimes water.74
To understand why these choices were made, we need to consider some other
reactions which the chemists participating in the dispute hoped to analyze.
Three important types of reduction experiments were performed and dis-
cussed by the main protagonists. None would have disputed the existence of
all the types, but the assignment of particular experiments to the categories
involved important controversial elements. From the mid-1770s on, Priest-
ley's classic experiment on the red calx of mercury served as an exemplar of
one type of reduction:
However, calxes that could not be decomposed through the gentle heating
employed in this first type of reduction could be reduced in the presence of
inflammable air (hydrogen). Thus
Finally, there was the classic method of reducing calxes, long known to smiths
and other metallurgists
How could one make sense of these reactions, either on Lavoisier's approach
or on the basis of commitment to [6b]?
Notice first that, prior to the understanding of the composition of water,
74. On Kirwan's views, whether phlogiston combines with vital air to form water or
fixed air depends on the temperature. At high heats, Kirwan agrees that water is always
formed, but " . . . it cannot fairly be inferred that water results from their union in any
lower heat; on the contrary, it appears that another compound of both, viz. fixed air, is
then formed" (1782; 1789 43).
284 The Advancement of Science
Lavoisier faced a serious difficulty with respect to (Rii). The initial description
of reactions of this type was not, in fact, as I have given it. The "dew" that
formed on the interior of the vessel was ignored, and chemists reported what
they had found as
But for the need to concede that something is absorbed in calcination, that
is to accept [4], such experiments would have been perfect for the phlogis-
tonians: if one takes a metal calx to consist of metal - phlogiston, then, by
identifying phlogiston with inflammable air, experiments of type (Rii') are
immediately accounted for. Nevertheless, even after the phlogiston theory
has been modified to accept [4], [6], and [6b], phlogistonians can account for
these reactions by supposing that inflammable air is phlogiston; that the phlo-
giston unites to the calx, restoring the metal and releasing the compound Z.
Even if this compound cannot be traced, it can be supposed that the phlogiston
is enough to restore the metal, even if it does not emerge in its pure state
but rather combined with Z (thus the resultant metals would be treated as
having impurities). By contrast, Lavoisier is committed to supposing that the
calx loses its vital air in the reduction and that both this and the inflammable
air somehow disappear.
After a careful search for the products of the reaction reveals the presence
of water, and after the acknowledgment of
Kirwan's version of the phlogistonian approach tries to combine [8a] and [8b],
supposing that inflammable air and vital air produce water at high heat and
fixed air at lower temperatures. Consequently, he holds that those calxes
formed at lower temperatures contain fixed air, while those that are only
formed at higher temperatures contain water.
We can now recognize three different approaches to explaining reduction
reactions. On Lavoisier's story, (Ri) reactions simply involve the decompo-
75. Kirwan goes on to refer to another experiment in which he believes that Lavoisier's
explanation will not work. In this instance, the heat involved was lower, and Kirwan
expresses skepticism about the possibility of forming water from inflammable and vital airs
at these lower temperatures. Lavoisier counters Kirwan's claims about reactions of these
gases at lower temperatures; see Lavoisier in Kirwan (1789 57).
286 The Advancement of Science
sition of a calx into its constituents, the metal and vital air; (Rii) reactions
involve the combination of inflammable air with the vital air of the calx to
produce water and leave the metal. In all cases where carbon is present, either
introduced as an observable reducing agent or contained as an impurity in
the metal, the vital air released combines with the carbon to produce fixed
air. Cavendish, advancing explanations based on [8a], must agree with La-
voisier's analysis of type (Riii), supposing that the production of fixed air
results from decomposition of the water in the calx, with the carbon uniting
with the vital air to produce fixed air and the phlogiston combining with and
thus restoring the metal. In (Ri) reactions, he will propose that phlogiston
from the water unites to the metal, releasing vital air, and, in (Rii) reactions,
the calx absorbs the phlogiston (inflammable air) releasing its water and
reviving the metal. Kirwan's account views some reactions as falling under
(Riv) and revealing the presence of fixed air in some calxes. Other calxes
contain water, and their reductions are treated as Cavendish would explain
them.
It should now be evident what Lavoisier had to do to defend the superiority
of his own scheme of interpretation. The suggestion that calxes contain fixed
air (one which Lavoisier himself had endorsed at an early stage of his tortuous
route to the approach of the 1780s) can only be turned back by providing
clear experimental evidence of the presence of residual carbon in those me-
tallic calxes that give rise, under reduction, to fixed air. Cavendish's preferred
hypothesis, [8a], requires considerable ingenuity for Lavoisier to launch an
experimental critique. Weighing the reactants and products in experiments
of type (Ri) will be inconclusive. Lavoisier solves his problem by considering
a different mode of calcination.
The famous "gun barrel" experiment is performed by placing iron filings
in a gun barrel, heating to red heat, and passing water through the gun barrel.
The products consist of the black calx of iron and inflammable air. All parties
to the dispute can agree on the following description of the experiment:
Cavendish and other phlogistonians will suppose that what occurs in this
experiment is that the inflammable air (phlogiston) is released from the iron
to leave the "basis of the iron" (iron phlogiston), and that this basis
combines with water to yield the black calx. Lavoisier proposes that the water
is decomposed to form vital air, which combines with the iron to form the
calx, and inflammable air, which is released. At a purely qualitative level,
nothing tells for or against either of these interpretations.
However, Lavoisier was able to show that an amount of oxygen whose
weight equaled the gain in weight of the calx would combine with the inflam-
mable air collected in the experiment to yield water, and that the weight of
this water is the weight of the water lost in the experiment. He could thus
provide an instantiation of CHEMICAL COMBINATION that would explain the
The Experimental Philosophy 287
one for which he and his colleagues can account. It is, of course, a consequence
of the explanation offered that pure metallic calxes will yield no fixed air on
reduction. In the concluding section of the French reply to Kirwan, Guyton
de Morveau elaborates the point by drawing on an experiment of Priestley's.
It is known that no carbonic acid is afforded in reductions, except when the
acidifying vital air is resumed by charcoal, and that its quantity when pro-
duced, is in proportion to the small quantity of charcoal which the vital air
accidentally meets with. The latter experiments of Dr. Priestley have put
these truths out of all doubt.76 He calcined iron in vital air by means of the
burning glass; he reduced it by means of very dry hydrogenous gas upon
very dry mercury; he found water in a quantity corresponding to the weight
which the iron had lost, and to the weight of gas absorbed; the remainder
was hydrogenous gas as before, and did not contain any fixed air at all. (In
Kirwan 1789 293)
These responses do not simply clear up the problem posed for Lavoisier's
chemistry by the existence of experiments which appear to be of type (Riv).
They directly challenge Kirwan's own account. For recall, the composition
of the calx is supposed to depend not on the types of precautions in preparing
reactants on which the French chemists insist but on the temperature at which
calcination occurs. If it can be shown (by experiments carried out by phlo-
gistonians, no less!) that variations in the production of fixed air are not
associated with the temperature of calcination, then Kirwan's attempt to
combine [8a] and [8b] runs into inconsistency.
I have traced one linealbeit an important onethrough the phlogiston-
ian controversy of the 1780s. We have seen how the general approach to
reasoning adopted in earlier sectionselimination of alternatives through the
production of inconsistency, forced retreats that open problems, employment
of background constraintsis concretely instantiated in the arguments of the
participants. Yet the story I have told is, for all its greater complexity than
Legend's tales of single crucial experiments, far too simple. The experiments
on calcination and reduction that I have considered were only a part of the
full body of experiments that both sides invoked. I claim that the rest of the
story is more of the same. Lavoisier and his colleagues successfully turn back
phlogistonian arguments, and they develop difficulties with phlogistonian ac-
counts of other types of reactions. I shall close this section with far briefer
discussions of these other aspects of the debate.
Acid-metal reactions took up just as much space in the chemical exchanges
as did the results about calcination and reduction. Participants agreed that
iron combines with nitrous acid (in water solution, to prevent a reaction that
Lavoisier describes as "too tumultuous" (1782 516)) to yield the calx of the
metal, together with nitrous air and water. On Lavoisier's interpretation,
nitrous acid is composed of nitrous air, water, and oxygen (vital air); the
reaction separates these constituents and joins the oxygen to the metal to
form the calx. Kirwan, committed to the notion that the calx formed contains
76. At this point, Guyton refers to (Priestley 1775 III 82).
The Experimental Philosophy 289
This scheme is not only more cumbersome than that proposed by Lavoisier
and his colleagues. Once again, Kirwan makes himself vulnerable by taking
on an unsolved problem. To make the accounting balance, he needs to identify
fixed air as a constituent of nitrous acid, thereby inviting requests to show
that nitrous acid contains fixed air (or carbonic acid). As in the case of
reduction reactions that yielded fixed air, the French chemists insist on pre-
cautions in preparing reactants. Admitting that small amounts of carbonic
acid can be discharged from nitrous acid at the beginning of reactions, Ber-
thollet notes that the reaction can be stopped so that pure acid is obtained:
All the nitre which is decomposed after this period, affords no carbonic acid;
the presence of this acid is therefore nothing but an accident, and we shall
hereafter explain why this accident takes place.
By following the different decompositions, we find, in them all, the two
principles we have acknowledged, and nothing else. (Berthollet in Kirwan
1789 118)
As with calcination and reduction reactions, Kirwan is squeezed. His approach
to calcination, founded on [8b], commits him to finding fixed air in nitrous
acid, and, when faced with Berthollet's results, he is again forced to
inconsistency.
By the mid-1780s, champions of phlogiston were clearly on the defensive.
Nevertheless, the controversy was not simply a matter of seeing whether their
framework could respond to a mounting number of difficulties. Kirwan, Cav-
endish, and Priestley try to show that Lavoisier's doctrine is problematic, and
part of the force of his case consists in the ability to turn back the challenges.
One important line of objection concerns the table of affinities which Lavoisier
had published in 1782. According to Kirwan, Lavoisier's account of the af-
finities of various substances for vital air is inconsistent with various experi-
mental results. Allegedly, substances that have lesser affinity for oxygen
(according to the table) are capable of combining with the oxygen attached
to a substance with greater affinity for oxygen (1782; 1789 41). Lavoisier
responds to this by acknowledging that his table of affinities is defective,
insisting that any such table has to be understood as valid for only one tem-
perature, and pointing out that the difficulties urged by Kirwan tell more
seriously against the phlogistonian approach (Lavoisier in Kirwan 1789 45-
55). As I mentioned earlier, the enunciation of principles of affinity that will
cover all the instances of chemical combination is an unsolved problem for
all chemists of the period, and Lavoisier contents himself with showing that
he can do at least as well as his rivals.
290 The Advancement of Science
Anyone who reads through the exchanges among Lavoisier, Priestley,
Fourcroy, Guyton de Morveau, Kirwan, Cavendish, and Berthollet should
be struck by the extraordinary difficulty of keeping in mind all the constraints
assembled by experiment. Each interpretative move has consequences that
ramify in ways that are very hard to foresee: introduce X as a constituent of
Y and how will one deal with experiment 57? Lavoisier's own painful struggle
to fashion a coherent interpretative scheme has been chronicled by Holmes,
and it is worth closing by reflecting on the fact that a clear line of argumen-
tation, emerging from the social process of debate and exchange, might, at
an earlier stage, be cognitively impossible for the pioneering investigator who
initiates that process. Holmes notes that, in 1776, Lavoisier regarded himself
as discovering that respiration converts common air to fixed aireven though
he had recorded this fact in 1774 and 1775. Holmes writes:
If I have a quarrel with this exemplary analysis, it lies in the hint that the
mental trait was an idiosyncrasy of Lavoisier's. Rather, I suggest, it was an
endemic feature of the problem situation. Between 1772 and 1783, nobody
could see how to fit together all the constraints and to manage a coherent
interpretation. By chasing down numerous blind alleys (in ways that Holmes
describes beautifully) Lavoisier was ultimately able to produce a wide-ranging
argument, consisting of the logically simple steps characterized in my account
of scientific reasoning, taken across a domain that was initially unsurveyable.
That argument was appreciated by many of his contemporaries and (rightly!)
led them to abandon the phlogiston theory in favor of the new chemistry.
The entire episode serves historians and philosophers of science as a reminder
that there is no need to invoke underdetermination, shifting standards, or
conceptual incommensurability. To paraphrase one of Lavoisier's co-workers,
Nous n'avons pas besoin de ces hypotheses-Id. The actual arguments of sci-
entists are much more interesting.
The Experimental Philosophy 291
10. Worries
I shall conclude by looking, far too briefly, at some obvious concerns about
the account of individual reasoning that I have offered. These concerns derive
from three sources: first, the recognition that there are existing philosophical
accounts of individual reasoning that treat scientific decision making far more
precisely than I have been able to do; second, the possibility that my efforts
at responding to underdetermination arguments only tackle relatively easy
instances (albeit those that have figured in traditional thinking about scientific
change) and that a more severe challenge can be generated by considering
debates about the legitimacy of appealing to experimental evidence; third,
and perhaps most dangerous, the intuitive sense that the processes for mod-
ifying practice that I have described would be unlikely to generate the kind
of progress for which I have campaigned in Chapters 4 and 5.
77. This is the simplest version of Bayesianism, one which assumes that statements
induced by perception receive probability one. For more complex accounts that allow
"observation reports" to have probabilities less than one, see (Jeffrey 1965), and, for
discussion, (Field 1978, Howson and Urbach 1989, Earman 1992).
78. The most impressive example here is Earman's account of evidence for general
292 The Advancement of Science
Moreover, Bayesianism is clear, precise, and unified: there are results about
proper reasoning and there is a single perspective from which these results
flow. I anticipate the criticism that, with these virtues, Bayesianism is supe-
riorperhaps as a descriptive theory, perhaps as a normative accountto
the cumbersome approach of earlier sections.
However, the Bayesian perspective has important shortcomings. First,
there is good psychological evidence that people do not naturally engage in
Bayesian calculations (Tversky and Kahneman 1973, 1974, Nisbett and Ross
1980) and that explicit assignments of probability are sometimes quite dif-
ferent from relative intensities with which beliefs are held (see Goldman 1986
chapter 15, especially section 2).
However, this in no way undercuts the possibility that Bayesians might
offer an excellent normative account. There is some residual appeal in the
notion that the reasoning of scientists involves a fumbling grasp of cognitive
strategies that the Bayesian methodologist brings to light, and that scientific
reasoning would be improved if the heuristic procedures that approximate
Bayesian patterns of reasoning were replaced by their precise counterparts.
There is an important similarity between my commendation of eliminative
induction and the Bayesian suggestions.79 I conceive of scientists using back-
ground constraints to devise a space of hypotheses and using statements
generated from interactions with nature to eliminate all but one of the can-
didates. I allow that there may be residual doubtpossibly quite significant
about the constraints that are used in generating the space. The Bayesian will
represent the same process by supposing that there is an initial assignment
of nonzero probabilities to the hypotheses that comprises the space, together
with the assignment of a probability to the "catchall" hypothesis that asserts
the falsity of all hypotheses in the space. This adds to my representation both
an assessment of the relative merits of the rivals and a quantitative evaluation
of the possibility that the space has been wrongly constructed. As the work
of elimination proceeds, the Bayesian apparatus can record just how the
fortunes of the rival hypotheses are changing. So it would seem to refine the
strategy that I view scientists as adopting, thus supporting the view that actual
scientific reasoning is a fumbling approximation to something clearer and
more sophisticated.
My claim, however, is that the extra detail is both unnecessary and ar-
bitrary. Imagine two scientists reasoning about any of the questions that have
occupied us in this chapter: one follows the eliminative strategy that I have
recommended, the other pursues the Bayesian policy of assigning precise
probabilities to candidate hypotheses. In what ways is the reasoning of the
latter superior? Given the absence of constraints on prior probabilities, it is
quite possible that the Bayesian arrives at some extremely unintuitive initial
assignment, with the result that, when his colleague has come to accept the
relativity (see Earman 1992). Bayesians also struggle with the logical problem of old evi-
dence (see Glymour 1980, Garber 1982, Howson and Urbach 1989, Earman 1992).
79. This is noted in (Earman 1992).
The Experimental Philosophy 293
last remaining candidate, the Bayesian still assigns that hypothesis a very low
probability. To be sure, there are convergence theorems about the long run
but, as writers from Keynes on have pointedly remarked, we want to achieve
correct beliefs in the span of human lifetimes.
Critics of Bayesianism paint for us the picture of bizarre assignments of
prior probabilities whose effects cannot be overcome sufficiently quickly, but
it is important to recognize that the effects are symptoms of the problem.
The root difficulty is that one ought not to partition the space of candidates
in vast numbers of the ways for which the Bayesian allows. Furthermore,
having recognized this, one can also see that any Bayesian partitioning of
that space imposes an arbitrary and unmotivated structure. Only in special
cases can responsible assignments of probabilities be made.80
Bayesianism also has trouble as an account of scientific reasoning because
of its preoccupation with epistemically perfect situations, cases in which there
is no inconsistency between statements that the scientist has reason to accept,
no tensions that result from the demands that an open problem be brought
within the scope of the accepted schemata. Just as Bayesians need some
machinery for distinguishing reasonable from unreasonable prior probabili-
ties, so too they need an apparatus to cope with those instances in which
epistemic costs must be counted and escape trees explored. As we have seen
in the reasoning of Darwin and Lavoisier, the hard work of reasoning does
not consist in performing calculations of probabilities but in bringing into
surveyable form the consequences of pursuing various responses to incon-
sistencies and open problems. As noted previously (footnote 48), a Bayesian
might attempt to use the apparatus of decision theory to try to make more
abstract and precise my account of epistemic costs and the ways in which such
costs are generated. Yet, I suggest, any such account would be parasitic on
the kinds of investigations I have outlined in general, and begun in particular
instances. Without the notions I have introduced in discussing underdeter-
mination problems or some surrogates for them, Bayesian accounts will always
be remote from the intricate reasoning of scientific texts (let alone from the
complexes of processes that underlie such texts). Moreover, the decision-
theoretic apparatus would, I suspect, introduce a need for arbitrary precision,
unmotivated specifications of probabilities.
Nevertheless, there are at least two ways in which Bayesian conceptions
prove valuable. First, there are surely occasions on which scientists can use
non-Bayesian forms of reasoning to assign responsible probabilities to can-
didate hypotheses and on which they can then proceed by using Bayesian
80. Wesley Salmon has articulated a position which combines the use of Bayesian
conditionalization with the employment of plausibility arguments to set the prior proba-
bilities (1967,1982). He calls this "objective Bayesianism," and, in some scientific contexts,
it may provide a useful refinement of the approach to reasoning that I endorse in the
chapter. However it seems to me that such contexts are relatively rare. Philosophical
purposes may sometimes be served by idealizing scientific contexts and using objective
Bayesianism to sort out tricky methodological issues. My discussion of the role of optimality
models in evolution and ecology (1988) is an effort in this direction.
294 The Advancement of Science
81. Of course, there have been philosophical attempts, most notably those of Laudan
and Shapere, to provide views of scientific reasoning that would subvert relativist claims.
Although I believe that these attempts have yielded important insights, I think that they
have been unconvincing because of failure to reconstruct historical examples in some detail.
The general account of the earlier sections of this chapter is an effort to improve on Laudan
and Shapere; the discussions of Sections 8 and 9 try to do the necessary historical work.
The Experimental Philosophy 295
ignored cannot be sustained (Shapin and Schaffer 1985).82 Shapin and Schaffer
reveal Hobbes as a subtle thinker who simultaneously opposed the conception
of knowledge articulated by Boyle, the social arrangements for the acquisition
of knowledge adopted by the early Royal Society, the use of the air pump
as a tool for promoting knowledge, and the specific claims made by Boyle
about the "spring of the air." Hobbes succeeds in advancing serious criticisms,
especially of Boyle's apparently hyperpositivistic methodological views
(Shapin and Schaffer 1985 81ff.). Thus, in this instance, it is initially plausible
to claim that there is (Kuhnian) underdetermination, based on commitment
to different views of knowledge, and, ultimately, to different "forms of life,"
so that there was no possibility of resolving the dispute and thus endorsing
the practices of the new experimental science on the basis of superior
reasoning.83
Nevertheless, I think that Shapin and Schaffer only succeed in exposing
how complex and difficult it was to defend the central claims of the new
experimental science. Hobbes's arguments about the proper form of social
arrangement for obtaining knowledge raise questions about authority, testi-
mony, and trust which will occupy us in the next chapter.84 What I hope to
do here is to show how the approach to the reasoning of individual scientists
that I have offered in earlier sections enables us to see how to criticize the
claim that differences in conceptions of knowledge and method made the
debate between Boyle and Hobbes a standoff.
Shapin and Schaffer take for granted the general doctrine of underdeter-
mination, supposing that there is no way in which opponents could be forced
to concede Boyle's experimental results and that there are always ways of
avoiding Boyle's explanations of those results. The former point is made
forcefully by invoking Harry Collins's concept of the "experimenter's regress"
(Collins 1985) and connecting it to the historical material. Shapin and Schaffer
write:
The claims Boyle made about his phenomena could be turned into matters
of fact [i.e., accepted within the community: PK] by replication of the pump.
But then other experimenters had to be able to judge when such replication
82. For a characteristically forthright presentation of the traditional view about Hobbes,
see (Conant 1957 57).
83. The Wittgensteinian phrase "forms of life" is used by Shapin and Schaffer, whose
central thesis is that the debate involved not only alternative conceptions of knowledge
and of knowledge-gaining practice but also commitments to different ideas about the proper
form of society. It is interesting to note that Hobbes saw the issue in a similar fashion: his
own explanation for the dismissal of his ideas identifies "the hatred of Hobbes [odium
Hobbii]" (Shapin and Schaffer 1985 379), which we might suppose to descend from the
widespread opposition to his political theory. However, whether there is a tight connection
or merely guilt by association is a question worth exploring.
84. I shall not attempt to reconstruct the arguments between Hobbes and Boyle about
the appropriate way in which to organize cognitive labor. However, I believe that the
conceptual apparatus that I assemble in Chapter 8 can be used to resolve such issues, both
for historical debates and for present disputes.
296 The Advancement of Science
had been accomplished. The only way to do this was to use Boyle's phe-
nomena as calibrations of their own machines. (226)
They also note that, even if Hobbes were to accept the Boylean findings as
genuine, he couldand didreject the hypotheses that were supposed to
explain those findings.
He contended that, whatever hypothetical cause or state of nature Boyle
adduced to explain his experimentally produced phenomena, an alternative
and superior explanation could be proffered and was, in fact, already avail-
able. In particular, Hobbes stipulated that Boyle's explanations invoked vacu-
ism. Hobbes's alternatives proceeded from plenism. (111)
This second type of challenge to Boyle leads very directly to the methodo-
logical differences between the two protagonists. For Hobbes could argue
that any explanatory endeavor must deploy hypotheses that are "conceivable,
that is, not absurd" (in Shapin and Schaffer 1985 362), and that plenist ex-
planations stand out as superior on this score.
Shapin and Schaffer's conclusions here rest on the tacit supposition that
any consistent position is defensible: they do not explore the kinds of diffi-
culties that Hobbes might encounter in pursuing the lines of resistance here
attributed to him, because they recognize that there is a consistent "way of
going on" and they assume that there are no epistemic criteria that can be
invoked to make more fine-grained evaluations.85 The task of this chapter
and its predecessor has been to show first that there is a basis for such criteria
(the versions of the external standard) and second that some further standards
(which are, we hope, in accordance with this basis86) are actually deployed
within scientific debates. Before we concede that the Boyle-Hobbes debate
is a standoff, we ought to see whether a more developed account of individual
scientific reasoning will break the stalemate.
The issues involved are precisely those considered in earlier sections. Just
as Galileo's telescope and Galileo's phenomena, Lavoisier's techniques for
purifying reactants and Lavoisier's interpretations of experiments are bound
up together, so too the idea of a properly working air pump is intertwined
with Boyle's results. Strictly speaking, the epistemic threat here is not that
of regress but of circularity: can we find a way to legitimize the employment
of an instrument or a technique without presupposing the correctness of the
controversial results to which it gives rise? Our previous discussion of ex-
amples suggests a way of breaking in to the circle. One establishes the reli-
ability of an instrument (or technique) by connecting its performance to
procedures that can be carried out independently; one shows the dependence
of particular variations in the performance of the instrument on changes in
the design or manufacture which can be understood by deploying indepen-
dently accepted schemata. Lavoisier and his colleagues could show how dif-
85. More exactly, they believe that the only such criteria are "local," so that while
discriminations can be made by those who are prepared to commit themselves in particular
ways, clashes among rival forms of life cannot be settled by the invocation of such criteria.
86. The third and final subsection will take up the worry that this is mere hope.
The Experimental Philosophy 297
to have favored an ability to think about social situations and relations. Ap-
parently we are far more successful at solving problems in reasoning when
their topics are socially significant than when they deal with features of the
asocial world (Cosmides and Tooby 1988).
It would be wrong to conclude that the appeal to Darwin actually reinforces
skepticism. Our current knowledge of human cognitive evolution is too ru-
dimentary to allow for much more than speculation. Instead of arriving at
conclusions on the basis of analyses that have more kinship with Kipling than
with Darwin, the appropriate response is neither optimism nor pessimism but
agnosticism. Our ignorance is, however, remediable, and we can hope that
more detailed accounts of human cognitive abilities and comparative studies
of related organisms might adjudicate this form of skeptical question.
Yet whether matters turn out well or badly on the question of our initial
state, the crucial issue surely concerns the possibility of successful correction
of practice. How much improvement might we expect from continually de-
ploying the strategies discussed in this chapter? Consider, for example, the
use of eliminative induction. Here the relevant skeptical worry focuses on
our aptitude for framing inductive problems. We might succeed in reviewing
what we take to be the relevant variety underlying an inductive generalization,
eliminating all the hypotheses we represent as rivals, and still be wrong be-
cause of the inadequacies of our ideas about possible hypotheses and de-
pendencies in nature. The right response to this kind of skeptical concern
seems to me to be prefigured in my discussion of the "pessimistic induction
from the history of science." First, there is the clear possibility that we are
relatively good at framing inductive problems with respect to some types of
phenomena and rather bad at coping with others. Through psychological
investigations and scrutiny of the historical record of the sciences, we can
even hope to draw useful distinctions. Second, even before engaging in de-
tailed studies of particular areas and inductive practices, we have ordinary
empirical grounds for thinking that the varieties within our ordinary classi-
ficatory groups are well understood: our success in comprehending the kinds
of factors that affect the behavior of metals, the colors of birds, the activities
of members of our own species is reflected in our own ability to produce
variation and in nature's apparent inability to surprise us with novel variants.
However, there is a far more serious skeptical worry generated by the
thought that the history of science is filled with debates that could not have
been settled by deploying the types of reasoning that I have canvassed. Skep-
tics contend that, at a number of times in the historical process that has led
to current science, there have been alternative possibilities for modifying
practice, each of which might with equal justice have been adopted. We can
imagine several possible histories of science, yielding divergent conceptions
of nature and rival sets of canons of reasoning, in each of which the protag-
onists retrospectively praise past decisions as exercises in self-correction. Be-
cause nothing distinguishes the actual course of events from these potential
histories, there is no basis for concluding that the actual evolution of science
is self-correcting while the others are not.
302 The Advancement of Science
1. Introduction
The general problem of social epistemology, as I conceive it, is to identify
the properties of epistemically well-designed social systems, that is, to specify
the conditions under which a group of individuals, operating according to
various rules for modifying their individual practices, succeed, through their
interactions, in generating a progressive sequence of consensus practices.
According to this conception, social structures are viewed as relations among
individuals: thus my departure from the tradition of epistemological theorizing
remains relatively conservative.1 If we remind ourselves of the framework
introduced in Chapter 3, then the general problem can be resolved into more
specific instances. If people join the scientific community through the social-
ization process I have sketched; if they form their divergent individual prac-
tices partly through interacting with nature and thinking by themselves, partly
by borrowing from and lending to others; if what is transmitted to the next
generation is shaped by their individual decisions and by their interactions,
how will the whole system best work to promote a progressive sequence of
consensus practices? Is it possible that the individual ways of responding to
nature matter far less than the coordination, cooperation, and competition
that go on among the individuals?
In what follows, I shall focus on two clusters of problems. The first set is
concerned with scientists' responses to others. The second considers the effects
of individual efforts on communitywide belief. So, in the earlier sections of
this chapter, I shall look at action on the individual. Later sections will study
the effects that stem from the doings of the individuals who make up a scientific
community.
1. More radical suggestions for socializing epistemology can be found in (Fuller 1988),
(Rouse 1987), (Longino 1990), and the writings of Bruno Latour. Many of the previous
chapters tackle the ideas and arguments that lead these writers to break with traditional
thinking about human knowledge. A different line of objection, posed forcefully to me by
Steven Shapin, is that the species of methodological individualism that I deploy in artic-
ulating my version of social epistemology cannot be sustained. For the moment, I shall rest
content with challenging those who believe that my individualistic framework is too narrow
to offer examples of social aspects of knowledge that cannot be accommodated within it.
303
304 The Advancement of Science
2. Authority
Reliance on authority affects all our cognitive lives, and, for present purposes,
we can distinguish three ways in which it permeates the cognitive lives of
scientists. First, there is the general epistemic dependence on the past that
figures in everyone's early intellectual ontogeny. We absorb the lore of our
predecessors through the teaching of parents and other authorities. Second,
at the time of entry into the scientific community, novices endorse a com-
munitywide conception of legitimate epistemic authority. Certain people are
to be trusted to decide on certain issues, and the novice must accept whatever
agreements they reach on those issues. Third, during the course of individual
research, scientists interact with one another, adopting the claims made by
some of their colleagues, investigating the proposals of others, ignoring the
suggestions of yet others, when the claims, proposals, and suggestions in
question go beyond what is agreed upon by the pertinent community.
A major theme of the foregoing chapters is that the overarching kinds of
authority that fall under the first two types do not trap us in inevitable error.
My accounts of progress and of individual reasoning attempt to develop the
idea that we work our way free of the mistakes of earlier generations through
further encounters with nature. But if this optimistic picture is to be sustained,
it must be the case that the third type of attribution of authority does not
interfere with the process of self-correction, but works constructively to fur-
ther the community project. I shall be concerned with this type of attribution
of authority, the differential assessment of peers.
To bring the problem into focus it is worth recalling some episodes from
recent science. In the spring of 1989, two electrochemists, Stanley Pons and
Martin Fleischmann, held a celebrated press conference, at which they an-
nounced the possibility of obtaining cold fusion on a table top. Prior to the
The Organization of Cognitive Labor 307
announcement that claim was incredible: according to the consensus practice
of nuclear physics, fusion cannot be achieved at room temperatures in the
fashion that Pons and Fleischmann described. However, the payoff if Pons
and Fleischmann were right would be enormous, and immediately after the
press release laboratory telephones began to ring. Electrochemists knew and
respected Pons and Fleischmann and, accordingly, took their apparently in-
credible claim seriously. Outsiders from physics had typically never heard of
either Pons or Fleischmann. Many of the telephone calls they placed posed
the same questions: Who are Pons and Fleischmann? Can they be trusted?
When told of the high standing that Pons and Fleischmann had within the
electrochemical community, the interested physicists began to consider the
possibility that the outlandish finding might be right. So a significant ex-
penditure of scientific effort was begun, as numerous phsyicists and chemists
tried to replicate the Pons-Fleischmann experiment.
Contrast this situation with another. During the past twenty years, self-
styled creation scientists have made periodic announcements about the co-
presence of dinosaur and human tracks in the same strata and about the
existence of human artefacts which, when dated by standard techniques, yield
ages comparable to those of supposedly ancient rocks. Their pronouncements
challenge paleontology just as Pons and Fleischmann questioned our under-
standing of nuclear fusion. But Duane Gish, Henry Morris, and their col-
leagues at the Institute for Creation Research do not inspire the same
dedicated investigations. They catch the ear of the scientific community only
when they are able to threaten, only when they have demonstrated an ability
to influence legislators and publishers, making it necessary for scientists to
divert time from profitable research to the enterprise of rebutting creationism.
The differences between the two cases are readily traced to differences in
authority. Pons and Fleischmann had considerable authority among elec-
trochemists, and they obtained authority among physicists because they had
authority for some people who had authority for some physicists. Gish and
Morris have no authority among paleontologists, nor do they have authority
with anyone who has authority for paleontologists. However, though this
contrast may throw into relief some basic features of the attribution and
withholding of authority in science, more mundane illustrations are useful for
bringing other points to our attention.
Consider anyscientist working in anylab. During the course of a day's
work there will be numerous opportunities for relying on others. Some will
be taken; others will not. Particular parts of the day's project will be assigned
to technicians, graduate students, support staff. The scientist will typically
perform other tasks herself. In addition, there may well be special oppor-
tunities to redesign some aspect of the course of research. Perhaps a journal
arrives with an article which, if sound, would enable a time-consuming pro-
cedure to be abbreviated. Or a grant proposal, sent for review, may suggest
an alternative sequence of experiments. Or a new catalogue may offer a novel
version of a relevant piece of apparatus. All these opportunities, both those
pursued or dismissed in the quotidian assignment of jobs and the more special
308 The Advancement of Science
3. Cooperation
The most obvious advantages of deference to authority are that it enables
individual scientists to pursue their epistemic projects more rapidly and makes
feasible investigations that would be impossible for a single individual. Sup-
pose that a scientist is dedicated to a particular inquiry: the scientist's over-
riding concern is to bring this inquiry to a conclusion by discovering the true
answer to a particular question. In terms that I shall employ more system-
atically in later sections, this imagined scientist is a pure epistemic agent, one
for whom the primary goal is to reach an epistemically valuable state. I further
suppose that the scientist has total resources (time, energy, money) E and
needs k items of information. Let the cost of acquiring each directly be C,
the cost of acquiring each from an authority be c. The scientist's project is
individually impossible but cooperatively feasible just in case:
4. Of course, the function F cannot be entirely arbitrary, but must meet the conditions
on probability (thus, for example, taking only values in [0, 1]).
5. A cautionary note on notation. In the sections that follow, numerous probabilities
and other parameters will be introduced. Given a limited lexicon, the same symbols have
sometimes been assigned different interpretations in different sections. Although I have
tried to maintain uniformity in notation for related problems, the reader should not assume
that symbols employed in sections dealing with very different topics have the same meanings.
The Organization of Cognitive Labor 309
noted, I suppose that your decision is motivated by the desire to bring this
project to a correct conclusion. Then you should rely on authority if
Even when you are perfect (p = 0), if the costs of borrowing are negligible
(c = 0), you can tolerate a maximum error rate of C/E in your potential
authority, and C/E may be sizable if you would have to expend a lot of your
resources on acquiring the information directly.6
The case in which the probability function F is linear is intermediate
between two others. If a new piece of information would yield rapid initial
returns, then, while it is still sometimes worth borrowing from others, the
maximum tolerable error rate is decreased. By contrast, if you would have
to expend considerable efforts, once the information is acquired, in learning
how to use it most effectively in your project, then it will sometimes be worth
borrowing even from extremely unreliable sources.
Up to this point, I have been assuming that the scientist's decision is domi-
nated by a particular type of epistemic intention: the goal is that the individual
scientist will solve the problem at hand. It is possible to imagine an even more
epistemically devoted scientist, one who cares only that the problem be solved
and who is prepared to pool resources with others in a richer cooperative effort.
A scientist of this altruistic bent would engage in a different type of decision
making, surveying the scientific community to ensure that her own efforts
worked with those of others in advancing the community's understanding.
6. As noted in the text, the discussion proceeds on the assumption that the project is
relatively complex. If we relax this assumption, there are two subcases:
(a) The project is so straightforward that you can attain probability 1 of getting a solution
whether you borrow or not; m(E - C) > 1 (a fortiori, m(E - c) > 1); unsurprisingly,
the maximum tolerable error rate is now p; i.e., it is only worth borrowing from those who
are more reliable than you.
(b) You can achieve probability 1 of succeeding if you borrow, but not if you do not;
m(E c) > 1, m(E - C) < 1; now the maximum tolerable error rate is (1
m(E C)) + m(E - C)p; it is easy to see that this value is greater than p, so that, here,
it is worth borrowing from people who are less reliable than you are.
310 The Advancement of Science
When I consider cognitive diversity in later sections, I shall, in effect, be ex-
ploring the kinds of ideal distributions of effort that altruistic epistemically
pure agents would aim to achieve. For the moment, however, I am interested in
looking at the cooperative inclinations of a different type of agent, an epistem-
ically sullied agent, one who is driven not only by a desire to solve the problem,
but also by the quest for priority which Merton (1973) emphasizes.
Consider the predicament of an epistemically sullied scientist, X, engaged
in a research project, when some peer, Y, announces a result that would, if
correct, provide a way to simplify the investigation. X reasons as follows:
"Suppose I make use of ys result; there are two possibilities, Y is wrong or
Y is right; if Y is wrong, then there is no chance of my solving the problem;
if Y is right then I will have a probability F(E - c) of arriving at a solution;
my rivals, whose resources are equal to mine, who also borrow from Y will
have an equal chance of solving the problem; if n of us borrow from Y and
N - n do not, then the expected number of problem solvers is nF(E c)
+ (N n)(l p)F(E C), so that, given that I produce a solution, my
chances of being thefirstare 1/[nF(E - c) + (N - n)(1 - p)F(E - C)].
Conversely, suppose that I do not borrow from Y; again, there are two
possibilities, Y is wrong or Y is right; either way, I have a chance of getting
the correct result if I set out to get the needed piece of information myself,
and, given that I get the right result, then there's a chance of F(E C) that
I will produce a correct solution to the whole problem; if Y is right, and n
members of the community borrow from Y then the expected number of those
who arrive at a correct solution is nF(E c) + (N - n)(1 p)F(E C)
as before; if Y is wrong, then the expected number of competitors is (N - n)
(1 - p) F(E C); either way, my chance of being the first solver is inversely
as the expected number of solvers."
This, I suggest, is eminently sensible reasoning for someone whose goal
is to be the first solver of a scientific problem, a scientific entrepreneur. It
involves a number of important assumptions. First, my imagined X makes
no distinction of talent: all the N scientists who are engaged in trying to solve
the problem are envisaged as equally likely to succeed. One way of relaxing
this assumption would be to suppose that X reasons not about the actual
community, but about a "virtual community" in which talented scientists are
allowed to count more than once. (Imagine that a community of five sound
but undistinguished scientists and one superstar is treated as of size ten, with
the superstar being regarded as equivalent to five "regular" scientists.) Sec-
ond, X takes the probability that he, or any of the others, will achieve the
correct version of Y's result, if they do not borrow, to be the same value
1 - p. In fact, X need merely assume that his own error rate is the average
of those whom he takes to be in competition with him. Finally, X does not
allow for any kind of partial credit: there is no payoff for being second or for
showing that Y's result is indeed right.
At this stage I want to point to a moral that will become familiar later.
From the community perspective, it is likely that sullied scientists will do
better than the epistemically pure. This is because a pure community heads
The Organization of Cognitive Labor 311
toward cognitive uniformity: either all the members find it worth borrowing
from y or they do not. By contrast, as Technical Discussion 1 shows, in the
sullied community, there are ample opportunities for division of cognitive
labor. Some follow the strategy of aiming for a quick victory by borrowing
from Y; others work independently. In this way, the sullied community hedges
its bets.7 That is, intuitively, a good thing. Later, I shall try to supply some
arguments that would underwrite intuitions of this general type.
So far, I have assumed an unrealistic condition of symmetry among my
imaginary scientists. All are supposed to have the same resources at their
disposal, to assess the error rates of themselves and others in the same ways.
By introducing asymmetries into the treatment of the entrepreneur's predic-
ament, we make it easier for different members of the community to pursue
different strategies. This is readily evident in instances in which, for one
member of the community p is zero, while, for another, p > q. (Intuitively,
one is well qualified to pursue the type of work that would lead to acquisition
of the information Y promises, while the other is not expert in such matters.)
The latter will find it profitable to borrow from Y, while the former will only
borrow if the problem is sufficiently complex (F(E - C) is sufficiently low).
There is a different type of asymmetry, one stemming from differences in
resources. As Technical Discussion 1 shows, we can expect those who have
fewer resources sometimes to take far greater risks in borrowing information
than would be justified by a sober epistemic assessment. If the scientific
community is divided into those who have relatively large amounts of re-
sources and those who have less, then we may expect that the latter group
will include some members who pursue what seem to be rather unpromising
ways of solving problems, while the resource-rich adopt more conservative
strategies. As I have already intimated, it is likely that this type of cognitive
diversity is no bad thing from the community perspective.
Technical Discussion 1
7. So presumably would a community of epistemic altruists, for they would adjust their
behavior to what others were doing and attempt to maximize the community's epistemic
utility.
312 The Advancement of Science
If q becomes too large (the potential source is too unreliable), then the
numerator will be negative; in this case there will be no borrowing, and
everyone in the community will do the work of obtaining the information.
Unless q is zero, the numerator is always strictly less than the denominator;
hence, provided that the community is not too small, there will always be
some people who do not borrow. The condition for there to be some borrowers
is that the numerator should be positive, that is,
If doing the work oneself is very expensive in terms of resources, and if the
problem of applying the information is sufficiently complex, then F(E C)
will be small in comparison with F(E c), and, in these circumstances, very
high error rates in the potential source can be tolerated.
Suppose, for simplicity's sake, that F is linear provided 0 < x < K, and
that E < K (where K is a constant). Assume p = 0 (the potential users of
Y's information are perfectly reliable at obtaining such information by them-
selves) , that c = 0 (the costs of borrowing are negligible), and that C = rE
where 0 < r < 1. Then, if r > q, some members of the community will borrow
from Y, and there will be an internal equilibrium state at which the number
of borrowers is given by
The Organization of Cognitive Labor 313
Figure 8.2
Provided that q > 0 (Y is not perfectly reliable) and the community is suf-
ficiently large, there will be some members of the community who do the
work themselves.
Asymmetry in Resources
Imagine that there are two competing scientists X and Z, and that a third
scientist, Y, announces a result that, if correct, would simplify the project on
which both X and Z are engaged. Each of the competitors has two strategies:
borrow (B) and do the work independently (I). I shall suppose that both
assess their own error rates and each other's error rate at the same value p,
that they both evaluate Y's error rate as q, that the costs, C, c, are the same
for both, but they have different resources, E1 for X and E2 for Z. The payoffs
to each are dependent on what the other does, and X and Z are effectively
playing a two-person game. For each pairwise combination of strategies, the
payoffs are determined by considering decision trees that trace the possible
outcomes. Figure 8.2 is the tree for the combination (B, B).
Following this approach, we achieve the following payoff matrix:
8. Here I assume that Z has no chance of achieving the first solution if Y is correct.
Intuitively, X's head start from borrowing will make it impossible for Z to catch up. This
contrasts with the analysis offered earlier, in which I did allow for the possibility that a
nonborrower might be the first solver, even in instances where the borrowed result was
correct.
314 The Advancement of Science
Understanding the outcomes of the competitive interaction between X
and Z involves identifying the equilibria for this game. The major features
of the situation can be discerned from considering the special case in which
F is linear, with F(x) - x/E1 for 0 < x < E, p = 0, and c = 0. In this case,
the payoff matrix becomes:
It is not hard to show that there are instances in which both scientists prefer
to borrow (BB is an equilibrium), and instances in which both prefer to work
independently of Y (II is an equilibrium). The only instances in which the
combination (BI) (X borrows, Z works independently) is an equilibrium are
the trivial cases in which C = 0. However, it is possible for the scientists to
achieve the combination (IB) (X works independently, Z borrows). This
occurs if Y's error rate is substantial, but not too high; if Z's resources are
fairly small in comparison to those of X; and if the work required to obtain
the information independently would consume most of Z's resources.
4. Attributing Authority
The discussion of the last section rested on the idea that scientists are able
to assess the error rates of others, or, to put it more positively, that they are
able to judge how reliable other members of the community are. How are
such estimates made? How are authorities evaluated?
Sometimes, like instruments, potential authorities can be calibrated di-
rectly. We compare the output of an authority with our own opinions on
topics where there is overlap. This is a temptingly simple suggestion, but I
propose that, from the beginning, we think about authority in a broader way.
We are interested in a function a(X, Y) that measures Y's authority for
X, or, more exactly, X's assessment of the probability that what Y says will
be true.9 Here X is an individual scientist and Y may be another scientist, a
research team, a journal, a series of scientific monographs, or some other
composite entity on which scientists may potentially depend. (There are often
9. More exactly, we are concerned with X's assessments of the probability that an
arbitrary statement (belonging to a particular classe.g., reports about a particular kind
of topic) will be true, given simply that Y produced or endorsed it. As we shall see in Section
11, the probabilities you assign to some statements are computed partly on the basis of the
content of those statements and partly by consideration of the authority of those who
produce them. This is most evident in cases where erstwhile trustworthy people announce
findings that, given the prevailing body of beliefs, seem highly implausible.
The Organization of Cognitive Labor 315
serious questions about the entity whose authority is to be assessed: when an
unreliable scientist publishes in a highly prestigious journal, known for its
strict refereeing, the authority attributed to the article may be an amalgam
of the authority of scientist and journal.) Attributions of authority are rarely
uniform across topic. I shall assume, in what follows, that the assignments of
authority are relativized to a range of issues with respect to which the potential
authority's deliverances can be assigned the same reliability.
An idealized treatment of authority that includes obvious social factors
can proceed by breaking a scientist's credibility into two parts: there is un-
earned authority that stems from the scientist's social position (either within
the community of scientists or in the wider society), the type of authority that
arises from being associated with a major institution or from having been
trained by a prominent figure; this contrasts with earned authority, that cred-
ibility assigned by reflection on the scientist's performances or through con-
sideration of others' opinions of those performances. So I propose that we
consider communities of scientists in which individuals evaluate one another
in accordance with the equation
(where w(X, Y) is a weight function, whose value indicates the relative im-
portance X takes unearned authority to have in evaluating Y). I shall consider
different possibilities for weighing unearned authority (different values of w)
and different methods of computing earned authority (different measures of
ae.).
Is this approach to authority complete? Perhaps the assessment of the
reliability of others depends not just on such large social factors as prestige
within the community but on the personal relations between evaluator and
potential authority as well.10 This could easily be incorporated by amending
the basic equation to
5. Direct Calibration
Let us now explore the consequences of various ways of attributing authority
to others. The best hope for minimizing the effects of the social structure is
to suppose that members of the community always set w to be 0 and calculate
earned authority through direct calibration. Under these circumstances,
a(X, Y) is simply Y's truth ratio with respect to the sample of statements
about which X has an independent opinion: in effect, X uses the straight rule
to project the probability that Y's claims will be correct from the frequency
with which Y asserts the truth, by X's lights, within the class of Y's statements
available for appraisal by X.11 So, for example, if X believes p1, p2, - p3,
11. This is gross oversimplification of a complex practice. Notice first that if the entire
corpus of Y's public pronouncements is considered it is likely to contain so many banalities
that the truth ratio will automatically be very high. By relativizing to a range of issues on
which Y's pronouncements are considered, I hope to prevent this difficulty. Intuitively, X
looks to Y's original contributions on a particular topic: so, for example, in assessing Pons
and Fleischmann, chemists recalled their particular, original claims within electrochemistry
and asked how many of these had been subsequently validated. In practice, the complex
assertions that scientists make are often used by their peers in evaluations of authority by
effectively atomizing their published and circulated work. Whole papers, or even series of
papers, are identified with a single, central claim, and track records are judged by consid-
ering how many of these claims are right (from the perspective of the assessor).
However, other considerations enter in. Trustworthiness is often appraised by consid-
ering the experimental and technical skills of the person in question and, in some sciences,
inspecting the visual representations that are produced. Hence, even given the relativization
The Organization of Cognitive Labor 317
p4, and - ps, and Y is recognized by X to assert p3, p4, p5, p6, and
- p7, then a(X, Y) = 2/3.
Is this at all realistic? Do scientists ever engage in this type of computation?
I believe that they do. The physicists who telephoned their electrochemist
friends to ask about Pons and Fleischmann typically wanted to know how
frequently the defenders of cold fusion had reported correct results about
similar matters. The electrochemists who responded were often in a position
to provide a minireview of the records of both Pons and Fleischmann.12
The simplest type of case in which direct calibration functions involves
the reporting of experimental results. Here the difficulties with individuating
and counting the deliverances of the target (mentioned in note 11) can be
addressed relatively straightforwardly because we have a preferred way of
classifying the reports that are of interest: we want to know how frequently,
in giving experimental reports of such-and-such a kind, our potential infor-
mant gives correct reports. However, there are other contexts in which direct
calibration can also be used. So we might ask, of a prominent scientist who
has a history of making large claims about promising lines of research in a
field, how often her suggestions about which directions to follow have proved
fruitful. Here we discover that connection between authority and credit at
which I hinted in the last section. One way to obtain substantial credit within
a scientific community is to acquire the reputation for knowing "where the
field should be going," and this is sometimes done through making claims
about the important lines of research. As one's peers compare these claims
with their own judgments about what has proved valuable, they engage in
direct calibration that confers authority with respect to the development of
the field, and this type of authority is an important determinant of credit.13
Deference to authorities might affect a scientific field by blocking the
spread of new ideas within the community, so that proponents of heterodoxy
would ipso facto lose credibility. A simple, somewhat artificial, way to focus
this worry is to consider what I shall call the alliance-splitting problem.
Imagine that we have three scientists A, B, and C. C advocates a finding
that would challenge the accepted ideas of the community to which all three
belong. In terms of direct calibration, B and C are natural allies, in the sense
to topic that I introduce in my analysis, the idea of computing truth ratios oversimplifies
the practice of direct assessment. I use it here as a way of contrasting the evaluation of
track records with the attribution of authority on the basis of social position (itself an eqully
complicated business), and, for present purposes, oversimplification enables sharp presen-
tation of the issues arising from the contrast. (I am grateful to Michael DePaul for cor-
respondence in which he raised interesting questions about the actual appeal to track
records.)
12. Here I am indebted to an insightful presentation by Jan Talbot in a symposium on
cold fusion held at UCSD in the spring of 1989. Electrochemists, but not physicists, could
(and did) calibrate Pons and Fleischmann directly.
13. I am grateful to Ernan McMullin for bringing home to me the differences between
attributions of authority about experimental findings and attributions of authority about
more "theoretical" matters. The discussion of the text is intended to note the differences
rather than to engage in the important work of analyzing them.
318 The Advancement of Science
(R) If you have to make a decision with respect to p, you are unable
to make that decision through independent inquiry, and there are exactly
two potential authorities whose opinions about p conflict, then you should
follow the judgment of the person to whom you assign higher authority.
Then, plainly, B will endorse C's finding. Under these (admittedly artificial)
conditions, we achieve the traditional epistemologist's Utopia: social factors
play no important role and, in effect, B's individual reasoning underlies her
judgments.
Despite its obvious unreality, this is a useful point from which to begin,
for it provides a baseline against which the disturbing effects of appeals to
authority may be measured. When authority is measured by direct calibration,
other informants are essentially extensions of the individual who appraises
them. Let us now look at more interesting ways of assessing the authority of
others, using the alliance-splitting test as an assay for identifying the episte-
mological roles of various social relations.
6. Prestige Effects
The most obvious way in which the presence of authorities can break up
natural alliances is through prestige effects. Suppose that our community is
one in which everyone who has undergone a reputable training program is
always assigned a value of unearned authority above some threshold. People
who have been associated with privileged institutions are attributed greater
unearned authority. Under these conditions, strong natural alliances can be
broken.
Consider, once again, our three scientists A, B, and C. As before, C
announces a new, controversial finding and A dismisses it as flawed. The
public assertions on questions that concern the three are as follows:
The Organization of Cognitive Labor
Figure 8,3
As Figure 8.3 reveals, (s, r) has to lie in the triangle whose vertices are the
points (0, 1), (1, 1), (1, k). Intuitively, there are many ways in which A's
prestige can split even a perfect natural alliance: all that is required is that
the weights B assigns to unearned authority be sufficiently high and the thresh-
old value of unearned authority for a respectable member of the community
sufficiently low. Under these conditions, if B is in the predicament of deciding
the merits of C's finding subject to the constraints on the artificial decision
situation I have describedin particular using rule (R)then A's prestige
will block the acceptance of C's challenge to orthodoxy. As we might have
320 The Advancement of Science
expected, and as Bacon and Descartes feared, appeals to authority can help
preserve the status quo.14
So far there are few surprises. Matters become more intriguing as we add
detail.
7. Indirect Calibration
There are numerous people whose advice might be valuable for a scientist
but whom the scientist is unable to rate directly. When you need conclusions
outside your speciality, direct calibration of potential informants is unreliable,
perhaps even impossible, since the sample of statements available for com-
puting a truth ratio is small, possibly empty. Under such circumstances, at-
tributions of authority must proceed either by use of unearned authority or
by what I shall call indirect calibrationusing the judgments of those who
have already been assessed in evaluating others.15
I shall simplify the general situation by focusing on cases in which the
only type of indirect calibration uses scientists who are directly calibrated to
evaluate those who are indirectly calibrated (the maximum length of paths
in the directed graphs is 2) and by ignoring possibilities of mutual assessment
and iterated adjustments. So we can write the earned authority of Y for X
as
14. Careful qualification is needed here. As Richard Foley pointed out to me, the
algebraic approach that I adopt does not mandate conclusions that authority will inevitably
have a cramping, conservative effect. It is possible to apply my basic algebraic approach
to different problem situations, within which appeals to authority can aid the introduction
of new ideas into the community: simply suppose that those most likely to make innovations
are those with highest authority. I have chosen to apply the algebra within a particular
problem situation, that in which innovations are likely to be produced by mavericks with
low authority, because this seems to be a common scenario in the development of at least
some sciences and because it highlights the apparent problems with deference to authority.
As we saw at the end of Section 3, and as we shall see again later, the pressures to take
risks are strongest on those who are at the fringes of scientific communities. Hence it seems
to me that the alliance-splitting scenario serves as a useful assay for examining the effects
of authority.
15. I owe to Steven Shapin the observation that Michael Polanyi had already recognized
the importance of indirect assessment of other scientists. See (Polanyi 1958 217).
The Organization of Cognitive Labor 321
be very unconfident about their direct assessments of their fellows, which is,
of course, why indirect calibration is used in the first place.16
Let us now consider our earlier example. We have three scientists whose
publicly expressed beliefs are as follows:
Plainly, the value of a(B, C) is reduced from that obtained through direct
calibration (sk + (1 s)), unless B gives no weight to A's unearned authority
(r = 0).
Under what conditions will the alliance between B and C be split? Suppose
first that h = 1, the case in which B is confident in her own power to assess
C and, in consequence, the situation most favorable for resisting alliance
splitting. The appliance is split if
that is
As Figure 8.4 makes clear, the region of (s, r) points that correspond to
alliance splitting is now increased: in particular, even when s = 0 (unearned
authority is discounted in the evaluation of C), the alliance is split if r > 0.62.
At the opposite extreme, if B is extremely unconfident about her ability to
evaluate C directly (h = 0) then the alliance is split if r > sk. It follows that
if B assigns equal weight to unearned authority in evaluating A and C, and
C is a reputable, but not high-ranking, member of the community, then the
B-C alliance will be split.
The example I have been discussing reveals complete disagreement be-
tween the evaluator B and the putative authority A and perfect harmony
16. Here again, issues about the kinds of statements that are being assessed arise. In
using y to rate Z, a(X, Y) must be computed by considering Y's reliability in judgments
about the reliability of others with respect to some particular range of scientific topics. a(X,
X) is X's measure of X's ability to rate others on that range of topics. If X believes that
her beliefs about this topic are likely to be wrong, then she may assign herself quite a low
value for a(X, X).
322 The Advancement of Science
Figure 8.4
The alliance will thus be split if h < 0.9. Unless B is extraordinarily confident
of her direct assessments, indirect calibration alone, without invocation of
unearned authority, suffices to resist the challenges of the maverick C.
The Organization of Cognitive Labor 323
8. Backscratching
Let us assume that wi(X, Y) = wi(Y, X) = wi.17 Let us also suppose that
degrees of authority are adjusted through a number of "rounds," and that
the values of au and ae remain constant through the process. Using indices to
mark the rounds, we get:
That is:
The process could only be runaway if the right-hand side of this equation
were less than zero or greater than one. Assuming that w2 < 1, it is plain
17. This is a nontrivial assumption, for there is no reason to think that people involved
in mutual assessment would give the same weight to all three factors. However, it helps
to simplify the algebraic treatment of the problem, and I shall forego discussion of the
more general case here.
324 The Advancement of Science
that the right-hand side cannot be negative, and not hard to see that it cannot
exceed one. For we maximize the numerator by setting au(X, Y), au(Y, X),
ae(X, Y), and ae(Y, X) all equal to one. Thus the value of the numerator is
less than or equal to:
Hence the right-hand side cannot be larger than one. In consequence, there
will not be a runaway process, provided that w2 < 1.
I do not wish to claim that the approach I have chosen (which embodies
a highly artificial symmetry between X and Y) is the only possible way to
model the effects of backscratching. However, I think it is noteworthy that
some rather straightforward assumptions yield scenarios in which backscratch-
ing doesn't have the consequence that mutual admiration societies will protect
members' opinions against criticism.
Technical Discussion 2
This reveals clearly that the effects of backscratching depend on the assign-
ment of weight to unearned authority.
(like other authors as diverse as Hagstrom and Latour) sees as one of the
main motivations for devoting a large proportion of a human life to difficult,
technical problems, scientists have to publish and market their ideas: "With
the possibility of credit comes the possibility of blame. Scientists cannot spend
very much time checking the work of others if they are to make contributions.
They reserve checking for those findings that bear most closely on their own
research, chiefly those that threaten it" (394). Hull thus offers us the picture
of scientific entrepreneurs, operating within a credit economy.
As in Section 3, we suppose that X is engaged in some research project
and that Y introduces something that could be used to reduce X's question
to some simpler issue, if it were accurate. Should X adopt Y's innovation and
pursue the apparently simpler project? Assuming that X has no competition
with respect to this project and that the task is simply to receive credit by
discovering a solution that is accepted by others, we can represent X's situation
as in Figure 8.5.
This tree represents a simplified conception of the possible futures for a
scientific entrepreneur whose possible strategies are to take over Y's poten-
tially helpful result or to ignore it and continue with the present course of
research. The new version of the predicament differs from that of Section 3
in two respects: I have left out of consideration the pressures that may be set
up by the presence of competitors who are working on the same project, and
I have explicitly couched the outcomes in terms of others' reactions to what
X does rather than in terms of X's attaining a correct solution.
Writing the utility of 0; as ui, I shall make the following assumptions:
(c) u22 > u21 There is greater utility in achieving an apparent so-
lution without Y's help than with Y's help.
(d) u3 < u2l The utility of achieving an unchallenged solution is
greater than that of being challenged and successfully diverting
blame.
(e) u3 > u4 The utility of diverting blame is greater than that of
receiving blame.
(f) u4 < 0 Being blamed has negative utility.
(g) p1 > p2 Use of Y's idea would increase the chance of achieving
an apparent solution.
Technical Discussion 3
The credit assigned for original or partially borrowed success is directly pro-
portional to the probabilities of success given borrowing or independent work.
By (9.3), the first terms are equal and the ratio of the second terms is:
328 The Advancement of Science
By (9.2), this is 1. Hence, given our additional assumptions, (9.5) and (9.6)
represent the expected utilities of each strategy and take on the same value.
In the null situation, borrowing from Y is as good as but no better than
continuing with independent research.
Thinking about the null situation is valuable because it enables us to
recognize various scenarios which would realize Hull's expectations about the
behavior of scientists and various scenarios that would defeat those expec-
tations. Consider, first, deviations from (9.2)-(9.4) that would favor borrow-
ing from Y. Borrowing is preferable if we hold the constraints on the null
situation constant except for replacing one or more of (9.2)-(9.4) with its
counterpart:
The situations in which these hold are the inverses of those that support
Hull's conclusions.
Technical Discussion 4
It should not be too hard to see that there are many ways of satisfying or of
violating this inequality. I shall consider a few special scenarios that are
particularly interesting.
Suppose first that the measure of authority within the community, a*,
used in calculating the probability of diverting blame, r, is simply the unearned
authority of the target individual, and that the probabilities of resisting chal-
lenge are the minimum degrees of authority of the participants in the at-
tempted solution, where these are computed by giving zero weight to unearned
authority and using direct calibration. Then, if t(X,Y) is the truth ratio that
X ascribes to Y, we have:
But unless X would rate Y as perfectly reliable, q1 < 1; hence, the first term
is negative, and, because u4 < 0 (see (f)), so is the second. Hence the ine-
quality is unsatisfiable, and it cannot pay X to borrow from Y. This is the
case of the self-confident prole and the unreliable aristo.
Conversely, if X is a low-ranking individual within the community, then
she may have little to lose through blame if a solution to the problem in hand
is successfully challenged, so that u4 is small. Suppose that she is also uncon-
fident, taking the community's appraisal of her into account in assign-
ing authority to herself. Under these conditions it is possible that a(X, Y) >
a(X, X), so that (neglecting the term in the small u4) (10.5) reduces to:
which supposes that benefits received with or without help from another are
inversely proportional to the probabilities of success. It is natural to consider
replacing (10.3) with an expression of Merton's "Matthew effect" (1973 443-
447):
The Organization of Cognitive Labor 331
becomes
Suppose further that the costs of being blamed if the apparent solution fails
are proportional to one's prestige:
When k is small, the qi are computed according to (10.2), and overall authority
is measured by direct calibration, this can be further reduced to:
When kh is small (either because the costs of blame are not too large or
because the individuals are sufficiently low-ranking), and the qi are measured
through direct calibration (as with (i)), borrowing will be preferable if
Comparing this with (10.16), we see that the effect of the leveling of rank
has been to double the value that the quantity p, t(X, Y)/p2 must exceed:
332 The Advancement of Science
the high-ranking may borrow much more easily from the low-ranking, and,
when equals borrow from one another, twice as much reliability is demanded
from the source or the probability of success through borrowing must be twice
as high (or there must be some combination of circumstances that produces
the same effect).
Using similar reasoning with respect to the other combinations, we get the
following assignments of payoffs:
For (BI) (X borrows, Z works independently):
Figure 8.6
The Organization of Cognitive Labor 333
It is not hard to show that there are choices of the parameters u21, u22, a(X,
X), and so forth, that will yield equilibria at (BB) and (II). The interesting
question is whether the asymmetry in status can produce a situation in which
there is an asymmetric equilibrium.
To answer this question it will help to make some special assumptions,
consonant with but not determined by my qualitative description of the sit-
uation. Let us assume self-confidence for both scientists: a(X, X) = a(Z, Z)
= 1. Relative status in the community will be appraised by supposing that
a*(X) = 1, a*(Z) = 1/2. I shall assume that a(X, Y) = a(Z, Y) = r, where
1/2 < r < 1, and that u22 = ku21 where k > 1. Normalizing the utilities so that
u21 = 1, the payoff matrix becomes
Recall from the end of Section 3 that differences in resources could yield
IB as an equilibrium, given appropriate relations among error rates and re-
sources. Here, however, given the assumptions made so far, IB cannot be an
equilibrium. For IB to be an equilibrium would require
These are jointly satisfiable provided that p2 > p1/(2 p1). However, the
constraints on k require that k - 1 be relatively small. Hence, we can obtain
an asymmetric equilibrium only when the relative chances of success through
working independently are substantial (so that there is a boost, but not an
enormous boost, from borrowing) and if the differences in payoff from work-
ing independently and using Y's result are relatively slight.
19. I shall relax this highly unrealistic assumption later in this section.
The Organization of Cognitive Labor 335
The more limited question I shall consider is, Under what conditions is R a
better strategy than both / and Ml
In advance of the finding, you would have assigned a probability of 0 to
the statement asserted by the alleged finder C. Now you use your assessment
of C's authority to grant a probability p that the statement is correct.20 You
also assign a probability q that the potential replicator will work reliably,
endorsing C's result if it is correct and rejecting it if it is false. The expected
utilities of the strategies can be written as
where v is the cost of ignoring the challenge (if true), w is the cost of
building on it (if false), u is the benefit of building on it (if true), and e is
the cost resulting from the loss of work that the replicator would otherwise
have done. I shall assume that u, v, and w are all appreciably larger than e,
and that w > u > v.
I is preferable to M if p < w/(u + v + w). Typically / will be preferable
to M because p will be small, p < 1/3, and, by the assumption that w > u >
v, w/(u + v + w) > 1/3. Moreover, if the replicator is highly reliable (q close
to 1), then R will be preferable to M even when p is relatively high. For R
is preferable to M if
This sets quite a strong condition on the reliability of the potential replicator.
Intuitively, if p is very small and if the costs of building on error are very
severe (w is large), the potential replicator had better be extremely reliable.
Furthermore, if the right-hand side is greater than 1, the inequality is unsatisfi-
able however reliable the potential replicator may be. This occurs if
20. As remarked in note 9, the probability is computed by considering two factors: the
content of the statement and the trustworthiness of the source. The weighting of the prior
probability (based on content, and, in the case at hand, 0) and the probability based on
utterance by the source (the reliability of the source on this range of topics) depend on
your assessment of the reliability of the two channels. How likely is it that the traditional
wisdom is mistaken? What is the chance that the assessment of the authority of the source
is suspect? The ratios of the probabilities assigned in answering these questions give the
relative weights assigned to the content-based probability and the authority-based proba-
bility. See p. 336.
336 The Advancement of Science
How should you, the philosopher-monarch, compute values of p and q?
Before C announced his finding you would have attributed to C a certain
authority a(C), to be bestowed on any result he asserted, and, on the basis
of your prior commitments, you would have assigned this particular result
the probability 0. The probability you now assign to C's finding should be a
weighted average of these, with the weights expressing your relative confi-
dence in your prior judgment about C and in the body of scientific lore that
underwrote your dismissal of any such result as that he has now announced.
So it is reasonable to write
where z is between 0 and 1, and is ever closer to 0 the more firmly entrenched
the parts of scientific practice that C's claim calls into question. I shall suppose
that q is simply your assessment of the authority of the potential replicator.
When well-established parts of science are called into question, it seems
reasonable to think that the past track record of those parts of science should
be weighted at least twice as heavily as judgments about the authority of a
single individual. Thus z should be less than 1/3, and, even given a perfect
potential replicator, I is preferable to M.
Consider now a community that meets the Pons conditions: (a) a(C) is
close to 1 (the person who announces the finding has high authority), (b)
there are potential replicators in the community for whom q is close to 1 and
e is negligible (people of high authority who could turn to replication without
serious cost to the community), (c) u and v are of the order of w/2 (there are
serious epistemic gains from incorporating the finding if it is correct). R will
be preferable to I if
Technical Discussion 5
The optimal community strategy depends upon the procedure for forming
consensus. Assume that the finding will be added to consensus practice just
in case r of those attempting to replicate it succeed in doing so (r need not
be independent of n: so, for example, community policy might be to accept
the finding if it is replicated by half those who attempt to do so, provided
that this number is at least two). The possible outcomes and their associated
probabilities can be represented as in Figure 8.7. Given that the finding will
Figure 8.7
338 The Advancement of Science
Assuming that the costs of losing the work of a single scientist in the quest
for replication are negligible, e = 0, then n should be chosen to satisfy
Having explored what the community would prefer to see happen in response
to a challenging finding, let us now consider whether scientists, attributing
authority in various ways and employing various decision rules, could work
their way to the type of strategy that the community favors. Imagine, then,
that you belong to a community of scientists and that some scientist, X, has
announced a surprisingly heterodox finding. I shall suppose that you have
two possible responses: you can ignore X and go on with your research project,
or you can attempt to replicate X's result, deferring your current project. It
is possible that there are conditions under which a third strategy, that of
incorporating X's finding into your own research work without any antecedent
check, might also become temptingfor reasons that we have explored in
earlier sectionsbut I shall suppose that these do not obtain in the present
case.
The strategy of ignoring X is basically a null option. By pursuing it you
do not change your future well-being, so that the extra utility that accrues to
you is 0. If you try to replicate, on the other hand, there are potential gains
and losses. If you are the first to replicate then there will be some benefit
u*possibly epistemic, possibly social, possibly mixed (the attainment of
truth, the acquisition of status, the recognition of having discovered something
important). If you are the first to show that X is wrong, then you will gain
v*, the utility accruing to the first refuter (again, this may be epistemic, social,
or mixed). But, by diverting your effort from the ongoing research project,
there will be costs of delay, d. On the basis of prior assessment of the authority
of X and prior reasons for assigning probability 0 to X's challenging finding,
you now suppose that there is a probability p that X is right. You are confident
that, in any attempt at replication, you will judge correctly. However, to
receive the benefits you have to be the first replicator or refuter. Your chances
depend on the extent of the competition. Suppose that you conceive of all
your colleagues as equally likely to win the race to replicate or refute. Then,
if n 1 join you in trying to replicate, the probability of your being the first
is 1/n. The expected extra utility of trying to replicate is therefore
Assume that the new finding is sufficiently important that it would gain
high credit for the first replicator and epistemic dividends for all researchers
in the field, so that u* > v*. If you and all your colleagues decide in the same
way, then at least one person will try to replicate X's result if
where Y is the scientist making the probability judgment and PF is the com-
posite entity Pons-Fleischmann. To obtain a nonzero probability, all that is
required is that nonzero weight be assigned to the prior authority of the
experimenters and that they have nonzero prior authority. All that was needed
from the telephone conversations on this score was the assurance that Pons
and Fleischmann were reputable, not outsiders like Gish and Morris.
The second issue is more subtle. Each scientist could assess for herself
the costs of delay in current research. The benefits of refuting a challenge to
orthodoxy depend, however, on the authority of the challengers within the
community. If nobody is prepared to defer to these challengers on this issue,
then, supposing that their challenge is wrong, little or no epistemic damage
will be done. Similarly, if the challengers are already seen as having low
authority, exposing the flaws in the present challenge will not redound to the
credit of the refuter. So v* will be a nondecreasing function of the perceived
authority of the challengers within the community.21 Suppose, for the sake
of simplicity, that
21. Michael DePaul pointed out to me the possibility that the utility of a successful
refutation might depend not only on the perceived authority of the challengers but also on
the number of people engaged in attempts to replicate. Thus, once the challenging finding
The Organization of Cognitive Labor 341
When telephone calls reveal that other scientists Y give a sizable value to
a(Y, PF), or when it is found that Pons and Fleischmann have high unearned
authority, scientists whose delay costs are small will make the decision to
investigate their finding.
Throughout earlier sections, we have faced the worry that appealing to
authority could retard desirable changes in science, making it difficult for
correct challenges to orthodoxy to be appreciated. Indeed, as we saw earlier,
reliance on authority in assessing the credibility of third parties can lead
scientists to take a jaundiced view of novel claims. I have explored a highly
idealized model in which the community preference is for some efforts at
replication, and I have suggested that a variety of epistemic or social forces
could lead scientists to modify their research in ways that satisfy this generic
preference. Moral: the impact of appeals to authority has to be evaluated by
recognizing the other forces that come into play in scientific decision making
even when a high weight is given to unearned authority, even when calibration
is indirect, a community is not doomed to stagnation if there are sufficient
rewards (social or epistemic) for entrepreneurs and critics.
But this moral only considers the problem in a relatively gross way. If we
have a community in which the new finding will be adopted into consensus
practice if just one person succeeds in replicating it, then, as we have seen,
it is quite likely that the pressures on individual scientists will lead someone
to attempt replication. As yet, however, we have no assurance that, in com-
munities with more complex consensus-forming mechanisms, enough member
scientists will attempt replication, or, in general, that the distribution of po-
tential replicators will bear any relationship to the community optimal dis-
tribution. Technical Discussion 6 explores the much harder problem of how
multiple replication might be achieved. It brings to light a serious difficulty.
is taken up, there is a possibility of bandwagon effects, even runaway bandwagons. This
can easily be modeled by supposing that
where v*, V, a* are as in the text, n is the number of scientists attempting to replicate/
refute the challenging finding, and k is a constant. Since my first concern is with the
conditions under which someone replicates, I am effectively looking at the decision problem
for someone when n = 0, so that there is no difference between the approach in the text
and that which would see utility as an increasing function of the number of other replicator/
refuters. The problem of adjusting the number of potential replicators will be considered
later.
342 The Advancement of Science
Technical Discussion 6
Figure 8.8
where
Suppose now that the community manages to reach the critical number
of r replicators.22 The number of those members of the community who
attempt to replicate will be given by the value of n such that
The community is well designed only if this value of n is close to nopt, computed
in the last section.23 More pointedly, we can say that a system of rewards for
member scientistsways of assigning values to u* and v*is epistemically
well designed if it produces coincidence (approximate coincidence) between
the value of n calculated here and nopt. It should now be clear that it is in
principle possible to assess the design of actual scientific communities with
respect to the replication problem. For we could investigate their method of
consensus formation; use it to compute the value of nopt for variable p, u, v,
and w; identify the ideal relationship between these parameters and u*, v*,
and d, and then examine whether or not this relationship is borne out on
actual occasions in which the replication problem arises. This would require
the distribution of nopt replicators to be stable when attained, and to be at-
tainable. Satisfaction of (12.8) only entails that the community will remain at
the optimum if it reaches it. It does not guarantee that the community can
reach the optimum.
24. Kuhn (1962/70 158-159) points out the importance for the scientific community of
persistent opposition, and his (1977) advances the general idea of the value of cognitive
diversity. Much of Feyerabend's earlier work is also devoted to celebrating the desirability
of keeping minority approaches alive, using them to sharpen both the dominant views and
one another (see Feyerabend 1963b, 1965, 1970). The root idea goes back at least to Mill
(1859).
25. I am grateful to Kim Sterelny for this observation.
26. This will not occur if the scientists are altruistic. If they are preparedand able
to adjust their cognitive commitments so as best to serve the community, then, of course,
they will replicate the decision making that identifies the community optimum. Whether
there are, in fact, any scientists so devoted to the community cognitive project as to waive
any concerns for their own epistemic advancement in favor of adopting a minority approach
("because someone ought to do that") is a question I leave for the reader.
The Organization of Cognitive Labor 345
The following sections consider the desirability and possibility of cognitive
diversity in a number of idealized situations. The principal theme that I shall
pursue is that epistemically sullied communities often seem to fare relatively
well. While communities of isolated knowers who proceeded in the ways
commended by individualistic epistemologies would often be doomed to cog-
nitive uniformity, social pressures and types of motivations often regarded as
antipathetic to the progress of science will turn out to be helpful. However,
as with the discussions of authority, competition, and cooperation, the picture
is complex, and there is no substitute for examining the details.
Before I offer my analyses it is worth noting an alternative approach to
the maintenance of cognitive diversity, one whose principal emphasis is on
the psychology of individuals rather than on the competitive and cooperative
relations among individuals. Perhaps we can appeal to the idea of cognitive
variation among individuals to recognize the possibility that, faced with a
murky situation, some scientists will prefer one approach, others another.27
My own discussions of cognitive variation in earlier chapters should make it
apparent that I am not unsympathetic to this idea, but I do not think that it
can be the entire story about the maintenance of cognitive diversity. In the
first place, there are numerous instances in which a scientist's persistence in
the face of severe difficulties seems best explained by the characteristics of
the social systemits framework of rewards and sanctions, its competitive
pressuresrather than by cognitive differences. Owen's opposition to Darwin
and the persistence of both Murchison and De La Beche in defending their
rival interpretations of the Devon strata are caused, in significant part, by
the social positions and competitive relationships among the actors. Other
instances can readily be garnered from the historical examples I have discussed
earlier (see, in particular, Chapters 2, 6, and 7). Second, although cognitive
variation among individuals may play an important role once rival approaches
have established themselves as genuine contenders, the route to establishment
often passes through extremely unpromising terrain. Those who articulate
initially implausible alternatives are kept going, at least in part, by various
kinds of nonepistemic pressures: fascination with a particular idea, devotion
to a national tradition, desire to make a name for oneself. These kinds of
motivations can, I suggest, be clearly discerned in the long intellectual voyages
that led to Copernicus's De Revolutionibus, Lavoisier's Traite, and Darwin's
Origin.
Thus, when In p1 ln p2 > kN, the optimal distribution is to assign all the
available workers to method I. Notice, however, that even when method I
has more intrinsic promise than method II (p1 > p2), there is a range of
conditionswhen ln p1 - In p2< kNunder which the optimal distribution
is to divide the community. Intuitively, a genuine division of cognitive labor
would be best for the community if there is a large available work force (N
is large), or if the methods respond quickly to the injection of effort (k is not
too small), or if the difference in intrinsic promise between the methods is
not too great (p1 and p2 are fairly close). The inequality given previously
represents the ways in which tradeoffs are made.
The functions considered in the last paragraph have the property that the
rate of increase in p(n) is maximal when n is small. This idea may not be at
all realistic. Perhaps the chances of achieving an answer by following a given
method increase quite slowly at first, then go up rapidly once a critical mass
of workers has accumulated, and eventually increase very slowly as saturation
is approached. We can describe this behavior by mimicking the logistic growth
equation of population biology, supposing that the pi are given by
30. I originally thought that these constraints would apply in all cases. However, as
Stephen Stich pointed out to me, too many cooks may spoil the broth. Imagine, for example,
that the method involves observing some sensitive organisms and that crowding in the field
would disturb the organisms' normal behavior. A live example of this phenomenon is hinted
at by Shirley Strum (1987 197-198). However, in many instances, I suspect that the prob-
abilities do indeed increase monotonically with the number of workers.
348 The Advancement of Science
If the probabilities are given by these functions, then there are various
cases of interest, depending on the value of k. Provided that k < 1/2, it is
possible to realize the intrinsic prospects of both methods, so the optimal
distribution divides the work force. If 1/2 < k < 1, it is not hard to show that
the optimal value of n is less than kN. More precisely, n should be chosen
so that
fail to achieve this, since all of them follow one of the dictates of epistemic
purity that lead to the distribution (N, 0). Moreover, because the structure
of VIM can only be fathomed by method II, their inability, as a community,
to hedge their bets is costly.
By contrast, in a neighboring nation, the chemical community is com-
posed of ruthless egoists. Each of the members of this community makes
decisions rationally, in the sense that actions are chosen to maximize the
chances of achieving goals, but the goals are personal rather than epistemic.
Those who elect to work on VIM do so because they believe that whoever
discovers the structure of VIM will win a prize coveted by all. They make
the simplifying (but not altogether implausible) assumption that, if a method
succeeds, then each person pursuing that method has an equal chance of
winning the prize. How should we expect the sullied community to distribute
its effort?
Imagine that the community has reached a distribution (n, N n). You
are a scientist currently working on method I, and you ponder the possibility
of switching to method II. The change would be good for yougiven my
assumption about your interests and aspirationsif it would increase the
probability that you win the prize. Now the probability of your winning is
the probability that someone in your group wins, divided by the number of
group members. (Intuitively, by choosing a method you buy into a lottery
that has a probability of paying up, a probability dependent on the number
of ticketholders; your chance of collecting anything is the probability that the
lottery pays up divided by the number of tickets.) Thus at (n, N n), it will
behoove a scientist working on method I to switch to method II if
includes all the optimal distributions, but it is quite possible that the com-
munity should get stuck at a suboptimal distribution, particularly at the
extreme (N, 0). Intuitively, if p2 is too small or N too large, there may be
no benefit in a maverick's abandoning method I.32 The good news is that
there are some instances of this general type in which the sullied community
not only does better than its high-minded cousins but actually achieves a
stable optimal division of cognitive labor. The bad news is that when the
community is too big, self-interest leads it to the same suboptimal state as
purity.
However, there is a remedy. The trouble with large communities (more
exactly communities for which kN is too big) is that a single deserter from
method I cannot contribute enough effort to method II to make that method
profitable. Several people need to jump ship together. Imagine, then, that
the community is divided into fiefdoms (laboratories) and that when the local
chief (the lab director) decides to switch, the local peasantry (the graduate
students) move too. Suppose that each lab contains q members and that the
director can thus bring it about that x members of the community switch
where x is less than or equal to q. Of course, if q > kN, then a single laboratory
can realize the intrinsic prospects of method II, and it is easy to see that there
are conditions under which a stable optimal distribution is attainable. If q <
kN, there are interesting special cases. Flexibility in the community is pro-
moted if q is greater than or equal to 3kN/4. Moreover, if q is at least
3kN/4, the stable optimal distribution is attainable if p 2 /p1 > 8k/9. Since we
are assuming that k is small, this result shows that the sullied community can
manage an optimal division of cognitive labor, even when the prospects of
method II are markedly less than those of method I. Moral: A certain amount
of local autocracylab directors who can control the allegiances of a number
of workerscan enable the community to be more flexible than it would be
otherwise.
My imaginary sullied community is, I think, a recognizable idealization
of a subcommunity depicted in James Watson's famous book The Double
Helix. If Watson's characters, Watson and Crick, are at one remove from
actual scientists, then the members of my imagined community are at a further
remove. However, the kinship is, I hope obvious: sullied communities may
come close to an optimal distribution because some enterprising members
see that it is to their advantage to try a method whose intrinsic prospects are
relatively low. In Watson's account, Watson and Crick take that kind of
gamble. The models I've constructed here might, whimsically, be called
"Lucky Jim models."
32. The condition for its being profitable to switch from method I to method II, if
everybody is using method I, is p2(3 - 2/kN) > p1k2N. Since, p2 < p1 (by our assumption
that method II is the inferior method), there's no chance of satisfying this if N is too large
specifically if Nk2 > 3. On the other hand, the stable distribution is attainable under some
conditions: (a) if Nk < 4 and p2(kN + 1)(kN - 2) > p1k3N2; (b) if Nk > 4, Nk2 < 3, and
p2(3kN - 2) > p1k 3 N 2 .
The Organization of Cognitive Labor 353
35. Here it would be more realistic to allow for the possibility that the community aims
to use the presently available theories in achieving a more adequate descendant theory that
would be closer to the truth than either of those now available. (See, for example, Martin
Rudwick's discussion of the great Devonian controversy in (Rudwick 1985), examined all
too briefly in Chapter 6, and in particular, the diagram at 412-413, which reveals the ways
in which later positions about the strata in question combined features of earlier theories.)
Obviously, allowing for the possibility of attaining closer approximations to epistemic goals
through synthesis of parts of current competitors would underscore the value of division
of cognitive labor.
The idea of a time of reckoning is a useful fiction, comparable to that employed by
population biologists in hypothesizing that the relative reproductive successes of different
types can be measured by looking at the value of a particular quantity (e.g., the number
of copulations achieved or the number of eggs fertilized) in a particular generation. In
principle, one should consider the total histories of consequences resulting from the choice
of particular strategies and the epistemic utilities associated with each. It is possible that
there are some instances in which it would be better to pursue a strategy of kind A, investing
all available resources in a currently more promising theory, and returning to its discarded
rival later, if the promise is not realized, and that the epistemic benefits of this strategy do
not show up on the simplified analysis I offer here. However, it is clear that there are other
conditions in which this is not so, and the simplification involved in positing a time of
reckoning does no harm.
The Organization of Cognitive Labor 355
pursued and if one overcomes its current anomalies, while the other does
not.
Now let us make the very optimistic assumption that nature, while not
forthcoming, is also not hostile: while correct theories may encounter an-
omalies, theories that successfully overcome all their anomalies are correct.
Thus, if we reach a conclusive state, then there is no need to worry about
false positive resultsin a conclusive state we resolve the issue and we resolve
it correctly. In light of this assumption, I shall assign epistemic utilities to the
outcomes as follows: if division of labor by following one of the B strategies
leads to a conclusive state, then we attain epistemic utility u2 (0 < u2 < u1);
if it does not, then the epistemic utility is 0 (we're still in the same predicament,
although our labor may have given us a clearer view of the problems that
each of the rivals faces).36
The expected utility of the more promising A strategy (assign all scientists
to T1) is easily computed. It is
To work out the expected utility of the strategy Bn (choose the distribution
(n, N n)), we need to recognize that a conclusive state in favor of T1 will
be attained just in case (a) T1 overcomes its current anomalies, (b) enough
workers are assigned to T2 to give T2 a chance. I shall assume (somewhat
arbitrarily) that the probability that (b) is the case is 1 if N n is larger than
some value m and 0 otherwise. The probability that (a) obtains is the prob-
ability that T1 is true multiplied by the probability that T1 responds to the
efforts of n scientists. Letting pi*(n) be the probability that a true theory Ti
responds to the assignment of n workers by overcoming its problems, we can
write the expected utility of Bn, where m < n < N m (so that both theories
are given a chance) as
36. My claims about the ui can easily be adjusted to reflect differences in views about
the values of particular outcomes, or even differences in specific situations that require
assignment of different values. Much more tricky is the task of replacing the hypotheses
about conclusive states with more realistic assumptions. In principle, one ought to allow
for the possibility that ingenuity can make a false theory continue to appear plausible, and
the idea of simple opposition between victory for one theory or an inconclusive dispute
should give way to study of the evolution of the probabilities assigned to the rival theories.
So, in effect, we want an estimate of the probability that each theory will be assigned if
there's a particular division of cognitive labor, and to use this to specify the probability of
making a correct decision at the time of reckoning. These complications will be explored
later in this section and in the following sections.
356 The Advancement of Science
At this point, it's necessary to make some assumptions about the parameter
m and the functions pi*. There are many ways to formulate the idea that both
theories should be given a chance: we could require that each theory must
be assigned sufficient workers to give it some high fraction of its maximal
chance of overcoming its problems (on the supposition that it is true), or that
the chances of each overcoming its problems be comparable (i.e., the prob-
ability that T1 successfully responds, given that T1, be roughly the same as
the probability that T2 successfully responds, given that T2), or that m be
simply some fixed relative frequency. I shall not try to choose among these
various ways of proceeding, for the reason that my assumptions about the
return functions pi* will allow for the existence of values of n that satisfy the
constraint, given any of the plausible choices for m.
As with the Lucky Jim models, I shall consider two possible kinds of
return functions: (a) pi*(n) = pi(1 - e- k n ), (b) pi*(n) = pi(3n2 - 2n3/
kN)/k2N2 (n < kN), pi*(n) = pi, otherwise. We can simplify the discussion
by supposing that p1 = p2, in both cases. For the return functions measure
the probability that the theory, if true, will be able to overcome its problems
if n workers are assigned to it, and it does not strain credulity to suppose
that these probabilities, for rival theories, might sometimes be the same
functions of n. In case (a), we may also assume that p = 1, that, given
the assignment of an infinite work force, the probability that a true theory
overcomes its problems is 1. In case (b), this supposition seems much more
questionable.
In light of the ideas of the last paragraph, some of the apparent implau-
sibility of my previous assumptions may be diminished. For, we can think of
the problem I am exploring in the following way: We want to compare the
merits of two community cognitive strategies, one that involves leaping with
the balance of the evidence, however slight, and one that involves patiently
waiting for a state that is genuinely epistemically conclusive before reaching
a decision. My idealizing hypothesis that, in an epistemically conclusive state,
the theory that has appeared to resolve all its problems is correct should
appear less outlandish if you think of the set of demands imposed for solving
all the problems as very rigorous. Under these circumstances, the probability
of our ever achieving the desired conclusive state may be quite small (a fact
that will be reflected in the choice of the parameter k that appears in the
return functions), so that the apparent optimism of the supposition that
choices made in conclusive states are correct is, in large measure, retracted
by recognizing that it is very difficult to reach such states.37
37. Of course, the point made in note 36, that we should think in terms of the evolution
of probabilities and of choices made at later stages as the probability of one theory comes
to seem much greater than that of its rival, remains correct. On the new way of looking
at things, there is an obvious class of strategies that are missing from the discussion: work
on both theories and transfer workers from one theory to its rival as the probabilistic
relations change. However, my depiction of the problem as a choice between two extreme
types of strategies has the merit that, if I can show that there are situations under which
one of the Bn's is preferable, it's clear that the optimal distribution involves a genuine
The Organization of Cognitive Labor 357
From the details of the algebra (see Technical Discussion 7) we can draw
an obvious moral. As in the case of the Lucky Jim models, there are specifiable
circumstances, albeit highly idealized, in which the distribution in an epis-
temically pure community diverges from the optimum and in which extra-
epistemic incentives bring the community to the optimal distribution. As in
the examples of Section 15, we can understand how it is possible that social
structures within the scientific community can work to the advantage of
the community epistemic projects by exploiting the personal motives of
individuals.
This is a useful beginning. However, the highly artificial assumptions that
have been made in my analysis foster the worry that my conclusions could
easily be subverted were we to proceed in a more realistic way. Moreover,
those assumptions also make it impossible to consider various factors and
combinations of factors that ought to be recognized. In consequence, an
extremely large part of the work of following sections will be devoted to
showing how we can do better. I shall offer a more refined model of division
of labor for theory choice, one that not only is more realistic but also provides
opportunities for introducing features that are absent from the relatively
simple Lucky Jim models.
Technical Discussion 7
Consider case (a), and make the simplifying assumption that p = 1. Then a
strategy Bn is superior to the more promising of the A strategies if
As with the Lucky Jim models, the optimal Bn is given when n = (kN +
lnq1 - Inq2)/2k. By our assumption that q1 and q2 are close in value, n will
be about 1/2, unless k is very smalland, if n is close to 1/2, both theories
have been given a chance. A sufficient condition for the optimal Bn to be
superior to the more promising of the A strategies is that
Intuitively, unless k is very small indeed, so that assigning half the available
workers to each rival would not provide a significant chance of achieving a
resolution of the issue, or if the utility of immediate action on the basis of
the true theory is very high, we can expect that an optimal distribution will
involve a genuine division of labor.
Turn now to case (b). Imagine, as with the Lucky Jim models, that k <
1/2, so that it would be possible to assign scientists to each theory in a way
that would give each a maximal run for its money. If n lies in the interval
division of laborfor that distribution is either given by the best of the Bn's or by one of
the missing strategies, all of which begin by dividing the labor.
358 The Advancement of Science
[kN, (1 - k)N], then the condition that both theories have been given a
chance is surely satisfied. So the crucial inequality for the preferability of one
of the Bn is
Thus, unless the maximal chances of a true theory's overcoming its problems
are low (p is small) or the utility of immediate action is high (u1 is large
relative to u2), the optimal distribution again involves a genuine division.38
Let us suppose that the important motive is each scientist's desire to be
singled out by posterity as an early champion of the accepted theory. If the
community is initially divided, with distribution (n, N - n), if the return
functions take the form (b), and if kN < n < (1 k)N, then there is a stable
attainable distribution, (n*, N n*), where
Provided only that k < q2 (a very weak assumption, given that k < 1/2 and
q2 is close to q1 (= 1 - q2)), this will be optimal.
38. I note explicitly that the distribution in an epistemically pure community will pre-
sumably be (N, 0). Since q1 > q2, pure scientists will choose T1 over T2.
The Organization of Cognitive Labor 359
Figure 8.9
too much loss of realism, we can keep matters simple by setting F1(n) = pn/
N. Using similar reasoning, F2(n), the probability that T2 will appear to solve
its problems given an assignment of n to T1, can be taken as (1 - p)(N
n)/N. G1(n) is the probability that if T1 solves its problems given an assignment
of n workers to T1 then T1 is correct. Now if T1 has had all the available
resources lavished upon it, then its appearance of success is far less impressive
than if it has done a lot with a little. We can use this intuitive idea to refine
the "give both a chance" constraint. Suppose that the status of T1 will be
effectively unchanged if all the available resources have been devoted to
articulating and defending it: that is, G1(N) = p. On the other hand, if T1
succeeds with no resources devoted to it, that is a considerable accomplish-
ment, and we can then take T1 to be maximally probable: thus, let G1(0) =
1 - a where a is small. If we suppose that the probability is a linear function
of n then we should set
By parity of reasoning:
Here we assume that the decision is made in favor of T1, the initially more
probable theory, and, since p > 1/2, U is positive. The expected utility of the
strategy that assigns n scientists to T1, N - n to T2, (n, N - n), is Vn where
360 The Advancement of Science
That is,
Technical Discussion 8
The expected utility of dividing the community (rN, (1 - r)N) is VrN, where
T(l/4, 3/4, p, r) decreases with p, so that, the closer the probability values
of the rivals, the more profitable it is to divide the labor rather than jumping
at once. Combining our results so far, we know that
Under these circumstances, it would be worth dividing the labor even when
T1 is three times as initially probable as T2.
The Organization of Cognitive Labor 363
Finally, if a = 0, b = 1, it is possible for work on the rivals to boost the
probability of one of them to 1, and there are no costs in delaying. As we
have already seen, if p = 1, there is no gain in dividing the labor. However,
if p < 1, we can always find r so that T(0, 1, p, r) > 0. Hence, if probability
1 is attainable, if delay costs nothing, and if the current value of the probability
of T1 is less than 1, dividing the labor is preferable to jumping at once.
Let us now turn to the second part of the task set at the beginning of this
section. How do we choose r so as to maximize V rN? I shall start with the
case in which a = 0, b = 1, and subsequently take a look at the effects of
increasing a, and decreasing p.
When a = 0, b = 1,
If then
That is,
p r* V0.5N Vr,N VN
0.6 0.55 0.52 0.53 0.2
0.7 0.6 0.58 0.6 0.4
0.75 0.62 0.625 0.65 0.5
0.8 0.64 0.68 0.71 0.6
0.9 0.68 0.82 0.84 0.8
0.95 0.69 0.905 0.92 0.9
I shall conclude by considering the effect of letting a depart from 0, (3
depart from 1. The general case is too complicated to provide a revealing
treatment, so I shall focus on the special case in which a = 0.1, b = 0.9.
Again, let r* be the optimal value of r (that which maximizes VrN). Substituting
values of p and using elementary calculus to find the associated maxima of
VrN, we obtain the following results40:
p r* V 0.5N Vr*N VN
0.6 0.59 0.38 0.39 0.19
0.7 0.68 0.44 0.47 0.38
0.75 0.74 0.48 0.53 0.48
0.8 0.81 0.53 0.6 0.59
The effect of increasing a and decreasing b has been to increase the optimal
value of r. Thus, when the maximal probability we can hope to obtain for a
successful theory is lower, and when the costs of delay are higher, the com-
munity division of labor should be less pronounced. We should note that, for
the special case under consideration, the strategy of setting r = p is a good
heuristic, and that, as in the earlier example, a fairly wide range of values of
r will yield acceptable results.
Technical Discussion 9
It is not difficult to show that the division generated by the idealistic com-
putation (henceforth the idealistic division) is preferable to jumping at once
in the special case in which a = 0, b = 1. For, setting a = 0, b = 1, r =
p, the inequality of the last section (17.4) becomes
which holds for all values of p. As noted in discussion of the other two special
cases ((a = 1/4, b = 3/4), (a = 0.1, b = 0.95)), there will be occasions on
which jumping at once is preferable to dividing the labor, and, on these
366 The Advancement of Science
occasions, performing the idealistic computation will lead the community to
a suboptimal strategy. (Thus, for example, if a = 1/4, b = 3/4, p = 0.6, the
idealistic division will be inferior to jumping at once.)
Perhaps actual scientists are neither thoroughly idealistic, believing that
truth will win out in the long run, nor thoroughly cynical, thinking that
apparent success, even that manufactured by the heavy investment of re-
sources, will determine community acceptance. Consider the possibility that
the estimation of the crucial probability, Prob(T1 will ultimately be ac-
cepted), is a result of a mixture of the two polar attitudes. Suppose then
that the members of the sullied community effectively compute the prob-
ability as
p r VrN Vmax
0.6 0.68 0.49 0.53
0.7 0.78 0.54 0.6
0.8 0.88 0.65 0.71
0.9 0.95 0.81 0.84
If the members of the community are more cynical, setting T = 3/4, then
the analogous results are
P r VrN Vmax
0.6 0.79 0.43 0.53
0.7 0.89 0.49 0.6
0.8 0.94 0.63 0.71
0.9 0.97 0.81 0.84
In a completely idealistic community, T = 0, we have
p VpN Vmax
0.6 0.52 0.53
0.7 0.58 0.6
The Organization of Cognitive Labor 367
Since, in the cases that interest us, VN is typically the minimum of Vr (1/2 <
r < 1), our mismatch measures the proportion of the maximal deviation from
the optimum that a particular distribution generates.41 It also has the advan-
tage that VN is the value achieved by the members of a completely cynical
community. The following table shows the values of the mismatch for various
combinations of p and T.
T = 0 0.5 0.75 1
p = 0.6 0.03 0.12 0.3 1
p = 0.7 0.1 0.3 0.55 1
p = 0.8 0.27 0.55 0.72 1
p = 0.9 0.5 0.75 0.75 1
Technical Discussion 10
Tradition Effects
Let the community of scientists be of size N (as in our previous examples),
divided into nt traditionalists, nr rebels, and N (nt + nr) neophytes. For a
traditionalist, the expected utility of defending T1 is
Prob(T1 is ultimately accepted)/(nt + Ornr + Ou(N nt - nr)).
The expected utility of defending T2 is
OtProb(T2 is ultimately accepted)/(nt + Ornr + Ou(N - nt - nr)).
For a rebel, on the other hand, the expected utilities are
OrProb(T1is ultimately accepted)/(nt + Ornr + Ou(N - nt - nr)) and
Prob(T2 is ultimately accepted)/(nt + Ornr + Ou(N nt nr)).
In the extreme case, members of a group associated with a rival theory
have no hope of receiving credit for work that leads to the establishment of
a theory. If this is so, then Ot = Or = 0. Neophytes are not treated so harshly,
and we can suppose that Ou = O where 0 < O < 1. Under these circumstances,
the decision problem for traditionalists and rebels is easy: traditionalists
choose T1 and rebels choose T2. Let us suppose that the neophytes distribute
themselves by using the idealistic computation, so that Prob(T1 is ultimately
accepted) = ProbI. Then neophytes choose T1 if
The first term here represents the standard frequency of T1 defenders within
the class of neophytes: if there were no division into rebels and traditionalists
in the community, then this would be the idealistic distribution. The second
term is a correction factor, representing the possibility that existing allegiances
The Organization of Cognitive Labor 369
make for an unbalanced distribution. From the view of the idealistic neo-
phytes, pnr ought to be (1 - p)nt. If nr is too large, then the neophytes will
correct by adding extra supporters to T1 (the correction term will be positive).
If nr is too small, then the complementary effect will occur, the correction
term will be negative, and the neophytes will attempt to rectify the situation
by disproportionately supporting T2.
But there may be too few neophytes. Consider a simple case, that in which
6 = 1/2. Now n* = pN + pnt + pnr - 2nt. This reaches its maximum value
of N nt nr when
If p exceeds this value then, do what they may, the neophytes will be unable
to restore the idealistic distribution across the community.
Let us now try to assess the effects of tradition on sullied communities
whose members use the idealistic computation. Without tradition, the number
of T1 supporters would be pN. With tradition, the number is
So defending T2 is preferable if
That is:
Compare this with a sullied community whose members use the cynical
computation, and in which there is a marked asymmetry in the awarding of
The Organization of Cognitive Labor 371
credit, t 1/N. Here the condition that there should be some support for T2
is
The equivalence of the two cases underscores a general moral: diversity within
a scientific community can be fostered in a number of different ways; although,
as we have seen, cynicism might lead to a condition of cognitive uniformity,
there are countervailing factors that can halt the potential bandwagon for the
dominant theory.
That is,
When m = 0, our "hybrid" agents are in fact epistemically pure, and the
condition reduces to the (unsatisfiable) p < 1/2. As m goes to infinity, the
agents are sullied, and the condition becomes p < N/(N + 1), which is the
condition for the representation of T2 in a sullied community.
The advantage of our new perspective is that it enables us to handle
intermediate cases. We can characterize these by the value of p, the maximum
value of p for which the community achieves some representation of T2. When
m = 1 (u/v = N ) , p 0 = (N + 2)/(N + 5); hence, even in a small community,
N = 10, T2 will achieve some representation unless p is close to 1 (p > 0.8).
Suppose, however, that our hybrid agents assign roughly equal value to truth
and credit for championing an accepted theory, u = v, m = 1/N. Then
p0 = 3N/(5N + 1); now, even in a large community, (N is infinite), p must
be less than 0.6 for T2 to be represented. Provided that their members are
not too pure, and provided that those members use the idealistic computation
of the probability of theory acceptance, communities of hybrid agents have
enough in common with sullied communities to achieve a tolerable division
of cognitive labor. However, if the scientists calculate the probability of theory
acceptance by using the cynical computation, then, as before, in the absence
of tradition effects (or other asymmetries), the result will be the uniform
distribution (N, 0).
Technical Discussion 11
If the community is very high-minded, and its members attach only very little
value to social recognition, then, of course, T1 and T2 must be very close in
perceived probability for T2 to be represented. If m = 1/2N then the maximum
value of p0 (as N becomes large) is approximately 0.55. If m = 1/10N (the
community members value truth 10 times as much as they care about credit),
then, even in a large community, the probabilities of the rivals must be very
close if T2 is to attract any representatives (the maximum value of p is about
0.512). Communities of more sullied scientists can obtain some representation
of T2 even when the probabilities are much further apart. If m = 2/N, the
maximum value of p0 is 2/3; if m = 4/N, then the maximum value of p0 is
3/4; and if m = 10/N, then the maximum value of p0 is about 0.86.
How much good does it do the community if its members assign some
weight to considerations of credit? Suppose that u = v, so that m = 1/N. As
we have seen, the condition for some representation of T2 is that p < 3N/
(5N + 1). Let us suppose that N = 10, p = 0.55. Then, solving the equilibrium
equation for r (= nIN), we obtain the value 0.83. The community optimum
value of r is about 0.5, so that, initially, we might conclude that the hybrid
community does rather poorly. However, when we compute values of VrN
for r = 0.5, 0.83, and 1, respectively, we discover that they are 0.5, 0.31,
374 The Advancement of Science
and 0.06. Thus, the expected utility of the distribution in the hybrid com-
munity, while lower than that obtained in a sullied community, is much closer
to it than the value obtained by the pure community.
42. Technical Discussion 12 does not show this, but application of the same methods
it employs readily demonstrates the point.
The Organization of Cognitive Labor 375
T1 Accepted T2 Accepted
Defend T1 kiu/n 0
Defend T2 0 u/(N - n)
where the ki may be greater or less than 1. Let us begin with the case in
which there are two subgroups, S1 and S2, whose members agree in assigning
a probability p to T1. Let k1 > 1, k2 < 1, so that S1's obtain greater rewards
from T1, S2's greater rewards from T2.
Suppose that the initial distribution is ((N1, 0), (0, N2)). T2 invades S1 just
in case
S11 and S22 are in cognitive harmony, in the sense that their views about which
theory is more probable accord with their ideas about which theory offers
greater incentives. By the same token, S12 and S21 suffer the correlative form
of cognitive dissonance.
Technical Discussion 12 shows that moving members of the extremal
groups, S11, S22, away from the theories to which both assignments of prob-
ability and incentives incline them is very difficult, so that support for both
theories will almost always obtain in the community. It also brings out the
possibility of an interaction effect between considerations of incentive and
considerations of plausibility that can diminish cognitive diversity. Differential
incentives and distinct assignments of probability appear to be independent
solutions to the problem of maintaining cognitive diversity, but it is possible
that, when they occur together, the beneficial effects of each can be cancelled.
Technical Discussion 12
Variations in Purity
Imagine that we have a community whose members are all hybrid agents but
who assign different values to m. Assort them into groups, S1, . . . , Sk with
376 The Advancement of Science
increasing values of m. Let the group sizes be N1, . . . ,Nk, and the hybridi-
zation parameters m1, . . . ,mk. Assume that, initially, all members of the
community defend T1. It will be easiest for T2 to gain representation in the
most sullied group, Sk. The condition for T2 to "invade" this group, given in
the last section, is
where N = Ni. Suppose that this condition is met. Then we can use the
equilibrium equation of the last section
That is,
Plainly, this condition is satisfiable only if N22 is very large. Similarly, if the
community reaches the distribution ((N11, 0), (N12, 0), (N 21 , 0), (0, N22)), then
T2 will invade S11 only if
The Organization of Cognitive Labor 377
That is,
That is,
That is,
Consonant with our initial assumptions, (21.11) and (21.13) are simulta-
neously satisfiable. It follows that it is possible for the distribution (a) to
collapse into (b): if (21.11) and (21.13) are both met, then it will pay one
member of S12 to switch to T2 and one member of S21 to switch to T1. The
conditions for other members of these groups to change allegiances will then
be just (21.11) and (21.13); so we should have sequential switching in each
group; if both groups have the same size (N12 = N 21 ) then, plainly, this can
lead to the replacement of (a) with (b), but, even if the sizes are different,
it is possible that the process should continue so that the changeover is
complete.
Let us now consider (b). Starting from (b), T1 would invade S12 just in
case
right-hand side of (21.15) is greater than 1/2. So (21.14) and (21.15) are
simultaneously satisfiable only if
This can easily be achieved provided that k1 is large and k2 is small. Hence
there are cases in which (b) collapses to (a).
Now consider that we begin from (b), the distribution that would be
favored by the epistemically pure. Is it possible that S12 should be invaded
and that S21 should resist invasion? To achieve this, we would require
Thus if k1 is sufficiently large (the incentives for S 12' s to defend T1 are very
strong) and if p is sufficiently large, the representation of T2 in the sullied
community is lower than the representation in the pure community. Here we
see an interaction between two types of forces that independently favor cog-
nitive diversity.
Figure 8.10
T1, champion T2, and explore further possibilities, respectively. I shall rep-
resent the possible outcomes by the tree in Figure 8.10.
In accordance with the discussion of Section 16, I shall suppose that F1(r1)
= p1r1, F2(r2) = p2r2, F3(r3) = p3r3. Relying on the treatment of the value
of exploring alternatives to the explicitly formulated hypotheses (see Chapter
7), I shall assume that Gi(r1, r2, r3) = pi (for i = 1, 2) if r3 < r*. If too little
attention is given to seeking rivals, then apparent success in solving problems
does nothing to boost the probability that a theory is true. However, the
value of r* ought to depend on the probabilities assigned to the extant rivals,
and I shall take r* = h(l {p1 + p2}), where h is a constant, less than one.
Appealing to the ideas of Section 16, let G1(r1, r2, r3) = ((1 a)r2 + p1r1)/
(r1 + r2), G2(r1, r2, r3) = ((1 - a)r1 + p 2 r 2 )/(r1 + r2), when r3> r*.
Without loss of generality, we can take T1 to be the more probable of the
two formulated rivals, p1 > p2. The utility of jumping at once by accepting
T1 is then
When r3 > r*, the expected utility of dividing the labor (r1, r2) is
Whence
380 The Advancement of Science
Quite evidently, the task of comparing the expected utilities of various di-
visions with the utility of jumping at once is far more complex in the present
case. Technical Discussion 13 explores the behavior of V r1,r2 by considering
a numerical example.
The numerical example only illustrates a possibility, revealing that, when
there are no significant costs of delay and when the probability of a theory
can be boosted close to one, it will sometimes be better for the community
not to jump to an immediate decision. How the resources are best assigned
to defense of each of the articulated rivals and to exploration of new options
varies with the parameter values, but, as noted, if a very high value is placed
on the emergence of new ideas, then it may be optimal to devote all the
resources to exploration.
Let us now turn to the decision problem for an individual scientist. Our
scientist may defend T1, defend T2, or explore new approaches. As in our
earlier discussions, we may suppose that defense of T1 would yield a return
of u if T1 were accepted and if the scientist were hailed as an early champion
of this theory. So we can take the expected utility of supporting T1 to be p1u/
n1 where n1 is the number of those in the community who devote themselves
to trying to show that T1 should be accepted. Similarly, we can take the
expected utility of defending T2 to be p2u/n2. Let q be the probability that no
accepted theory emerges, given that neither T1 nor T2 is accepted. Then, by
exploring, the scientist effectively buys into a lottery which will pay up only
if two conditions are met: first, neither T1 nor T2 must win acceptance; second,
some accepted theory must emerge from the process of exploration. Given
that these conditions are met, the scientist's chances of receiving the payoff
are inversely proportional to the number of explorers. The payoff for first
articulating a new option, ultimately accepted by the community, is, I assume,
large. Let the value be w, where w > u. The expected utility of exploring is
then
That is,
The left-hand side represents the chance that an accepted theory will emerge
from the exploratory process, given that neither T1 nor T2 is accepted. The
more confident the scientist is that failure of T1 and T2 will correspond to
success of one of the new ideas generated by the explorers, the more easy it
is to satisfy the inequality. But, given a fairly large community, exploration
is often worthwhile even if the scientist thinks that it yields only a small chance
of turning up an ultimately accepted theory. For example, if p1 + p2 < 1/2,
(1 - q) is required only to be larger than 1/N; if P1 + p2 < 3/4, (1 q)
must be bigger than 3/N. As Technical Discussion 13 shows, when it is over-
whelmingly likely that exploration will deliver a successful theory, provided
that the articulated candidates fail (q 0), the equilibrium number of ex-
plorers will be (1 (P1 + p2)). Even in the worst case, when w = u, if
q = 0, the pressures on individual scientists yield sufficiently many explorers
to give an acceptable distribution. We can have some confidence that moving
from the highly idealistic problem of 17 to the slightly more realistic version
considered here does not undermine our main conclusions.
Technical Discussion 13
Numerical Illustration
Let us suppose that a = 0, b = 1 (conditions, as we have seen, favorable
to dividing the labor). Let p1 = 1/2, p2 = 1/4, r1 = 1/2, r2 = 1/4, g = 1/2
(finding a new option is half as valuable as discovering the truth), and let
h < 1. Under these circumstances r3 (= 1/4) > r* (= h/4) so that our
expression for Vr1,r2 applies.
Straightforward substitution shows that
so that the distribution (1/2, 1/4, 1/4) is preferable to jumping at once. How-
ever, when the value of generating a new option is set so high, the optimal
strategy is to devote all of the work force to exploration
If q is highthat is, if there is a strong chance that exploring will not result
in any successful new theory, even if T1 and T2 failthen the equilibrium
number of explorers may be fairly small.
However, suppose that q = 0. Recall that, for the successes of T1 to
increase its probability, G1 must take a value greater than p1, and this requires
that r3 be larger than r* (< 1 - (p1 + p2)). If scientists are sure that, if T1
and T2 do not succeed, then something that emerges from exploration will
ultimately be accepted, then the community will contain enough explorers to
allow for serious increases in the probability of T1 or T2. At equilibrium,
r3 = 1 - (p1 + p2).
Technical Discussion 14
Suppose that there are N1 aristos and N2 proles in the community (Nl + N2
= N). Then, if there are n (<N1) aristos in the clique, the expected payoff
is
Suppose k = l.44 Let Q > 0.9. Then sufficient conditions for the community
to prefer a clique of aristos are:
44. This may be unrealistic if some proles are young members of the community who
may become aristos in time. For then the losses of their work due to delay are only a
fraction of the losses of the work of an aristo. But, comparing the total values of a lifetime's
work, the promising prole may be very close to the aristo. In that case, k might be signif-
icantly greater than 1, l approximately 1. I shall ignore this complication and other intricacies
that result from the age structuring of populations.
The Organization of Cognitive Labor 385
Since ln{2(l - Q)} < 0, dH/dn > 0 if lPe > kd. Thus, if the probability of
rebellion is high enough, or the costs of delay small enough, the community
would prefer to involve all the aristos in independent decision making. If,
however, lPe < kd, then the optimum number of aristos is given by
Even if d is tiny in comparison with u (u = 1010d, say), this will only give a
value of 10 for n, which, in a large community, may be a small proportion
of the aristos.
Elitism can become even more pronounced if we suppose that the costs
of delay increase with the number of independent decision makers. In more
democratic communities, even more time must be spent in articulating ar-
guments for others to appraise, and this will lengthen the period through
which the work of clique members is lost to the community. If, for example,
the delay cost for a prole is dn2, then, with the numerical values of the last
paragraph, the value of u/d must be 3 1012 for n to be as large as 10.
Another obvious amendment of my analysis is to suppose that the prob-
ability of rebellion varies inversely as the size of the clique. The decisions of
small, autocratic consensus-forming groups are likely to arouse resistance
among nonmembers. Suppose, then, that P = h/n. Under these conditions,
the expected utility of having a clique of n aristos is
We only obtain an elitist solution (that is: n < N1) if the right-hand side is
negative. If
which is approximated by
which can only occur if the members of the community are antiauthoritarian
(high h), or the community is small (low N), or the costs of delay are very
small in comparison to the value of a lifetime's work (d/e is very small).
The Organization of Cognitive Labor 387
24. Conclusions
As I noted in section 1, there are two main results of the foregoing analyses.
First, motives often dismissed as beyond the pale of scientific decision making
can, under a wide range of conditions, play a constructive role in the com-
munity's epistemic enterprise. Second, the details matter. The effect of various
types of factors (authority, elitism), which we might have thought of as acting
in a single fashion across scientific contexts, depends on features of the social
situation and of the decision problem.
These results flow from a formalbut highly idealizedtreatment of com-
munities and of individual decision-making. What can we achieve by modeling
scientific communities in so artificial a way? I have begun by employing an
unrealistic model of human decision making (who can give precise probabil-
ities? who has definite utility functions? who knows enough about others to
be able to carry out, even in principle, the computations I have labored
through?). I have proceeded by making numerous further assumptions in the
interests of algebraic simplification (think of the many occasions on which I
have regarded particular groups as homogeneous in some important respect).
These are defects that I would be happy to overcome. It is possible that a
more realistic approach to cognition could be formulated with sufficient pre-
cision to enable us to achieve clear results about the outcomes of decisions
made by the members of a community. It is possible that a more elegant
general approach would expose the crudities of my analyses and offer us a
more global perspective on the phenomena I have tried to study. Lacking
these potential advances, the discussions of the previous sections should be
viewed as first efforts, attempts to raise questions that I take to have been
slighted within epistemology and to devise concepts and tools for solving
them. I would be delighted were others to improve those concepts, or to
transform the tools, rendering my treatment of the problemsbut not the
problems themselves!obsolete.
Finally, I want to address the puzzled reader who regards this chapter as
part of a different book, one that does not contain its predecessors. Recall
some major claims of earlier chapters. Fields of science, I have suggested,
make progress in broadly cumulative fashion. As those fields progress, sci-
entists come to employ cognitive strategies that are superior, when judged
by versions of the external standard. Despite the existence of periods in which
388 The Advancement of Science
45. Here I deliberately use the language of (Latour 1987). However, it should be noted
that Latour's position is far more subtle than the view I sketch, and he would be as unsatisfied
as I am with my imagined polar opposite to Legend.
The Organization of Cognitive Labor 389
390
Envoi 391
scientific debates, there has been no general account on which philosophy of
the special sciences could draw. I hope that some of the concepts introduced
will prove useful in illuminating parts of scientific activityas, for example,
a predecessor of the notion of a practice was valuable to me in discussing the
growth of mathematics (Kitcher 1983a) and the structure of sociobiology
(Kitcher 1985b). In particular, I believe that the project of trying to map the
significant questions of a field can bring into focus areas of science in which
there are many rival ideas about the proper development of the field.
Finally, as I have remarked several times, the foregoing chapters leave
untouched some of the largest questions about science. Much of the opposition
to "scientism"the trendy castigation of "modernity" as bewitched by sci-
encerests on faulty views about scientific knowledge. Yet, even if the met-
amorphosis of Legend attempted here clears away those errors, it does not
address the issue of the value of science. To claim, as I have done, that the
sciences achieve certain epistemic goals that we rightly prize is not enough
for the practice of science might be disadvantageous to human well-being in
more direct, practical ways. A convincing account of practical progress will
depend ultimately on articulating an ideal of human flourishing against which
we can appraise various strategies for doing science. The extreme positions
are clear. At one pole, it is suggested that science, as practiced, is a terrible
thing, and that human beings should want none of it; at the other, that science,
as we have fashioned it, is already perfect. Neither extreme is likely to be
right.
In accordance with the approach taken in Chapter 8, we can envisage a
very general problem of optimization. Given an ideal of human flourishing,
how should we pursue our collective investigation of nature? Beyond my
attempt to understand the epistemic features of the scientific enterprise lies
this far broader question about science, a question that a critical philosophy
of science ought to address. It would, I believe, be highly surprising if any
of our institutions, evolved, as they have, by serendipitous routes, were to
be vindicated by a careful optimality analysis. Given a clear view of the
epistemic achievements and prospects of science, how should we modify the
institution so as to enhance human well-being? Reflective understanding and
constructive critique should, I believe, replace both sleepy complacency and
Luddite rage. The philosophers have ignored the social context of science.
The point, however, is to change it.
Bibliography
Allen, Garland. (1978a). Life Science in the Twentieth Century. Cambridge: Cambridge
University Press.
. (1978b). Thomas Hunt Morgan. Princeton: Princeton University Press.
Anderson, John. (1983). The Architecture of Cognition. Cambridge, MA: Harvard
University Press.
Appel, Toby. (1987). The Geoffroy-Cuvier Debate. New York: Oxford University
Press.
Aristotle. Physics in Sir David Ross (ed), (1966). The Works of Aristotle, volume 2.
Oxford: Oxford University Press.
. Posterior Analytics in Richard McKeon (ed), The Basic Works of Aristotle.
New York: Random House.
Barber, Bernard. (1952). Science and the Social Order. Glencoe, IL: Free Press.
Barnes, Barry. (1974). Scientific Knowledge and Sociological Theory. London:
Routledge.
. (1992). "Realism, Relativism, and Finitism," in Diederick Raven et al. (eds),
Cognitive Relativism and Social Science. New Brunswick, NJ: Transaction
Books, 131-147.
, and Bloor, David. (1982). "Relativism, Rationalism and the Sociology of
Knowledge," in M. Hollis and S. Lukes (eds), Rationality and Relativism.
Cambridge, MA: MIT Press, 21-47.
Beatty, John. (1985). "Speaking of Species: Darwin's Strategy," in (Kohn 1985), 265-
281.
. (1986). "Pluralism and Panselectionism," in P. Asquith and P. Kitcher (eds),
PSA 1984 Volume 2, East Lansing, MI: PSA, 113-128.
Benacerraf, Paul. (1973). "Mathematical Truth," Journal of Philosophy, 70, 661-679.
Bennett, Jonathan. (1966). Kant's Analytic. Cambridge: Cambridge University Press.
Bernstein, Richard. (1988). "The Rage against Reason," in (McMullin 1988) 189-
221.
Biagioli, Mario. (1990). "Galileo, the Emblem Maker," his, 81, 230-258.
Bishop, Michael. (forthcoming). "The Theory-Ladenness of Perception," manuscript.
Block, Ned. (ed) (1981). The Imagery Debate. Cambridge, MA: MIT Press.
Bloor, David. (1974). Knowledge and Social Imagery. London: Routledge.
Bogen, James, and Woodward, James. (1988). "Saving the Phenomena," Philosoph-
ical Review, 18, 303-352.
Born, Max, and Wolf, Emil. (1980). Principles of Optics. Oxford: Oxford University
Press.
Boscovich, Roger. (1763/1966). A Theory of Natural Philosophy. Cambridge, MA:
MIT Press.
392
Bibliography 393
Bowler, Peter. (1983). The Eclipse of Darwinism. Baltimore: Johns Hopkins Uni-
versity Press.
Boyd, Richard. (1973). "Realism, Underdetermination, and a Causal Theory of Evi-
dence," Nous, 7, 1-12.
Brandon, Robert. (1978). "Adaptation and Evolutionary Theory," Studies in the
History and Philosophy of Science, 9, 181-206.
, and Burian, Richard, (eds) (1984). The Units of Selection. Cambridge, MA:
MIT Press.
Brandt, Richard. (1959). Ethical Theory. Englewood Cliffs, NJ: Prentice-Hall.
Bromberger, Sylvain. (1963). "A Theory about the Theory of Theories and a Theory
about the Theory of Theory," in B. Baumrin (ed), Proceedings of the Delaware
Seminar. New York: Interscience.
. (1966). "Questions," Journal of Philosophy, 63, 597-606.
Buchwald, Jed. (1981). From Maxwell to Microphysics. Chicago: University of Chicago
Press.
. (1989). The Rise of the Wave Theory of Light. Chicago: University of Chicago
Press.
Burian, Richard. (1985). "Adaptation," in M. Grene (ed), Dimensions of Darwinism.
Cambridge: Cambridge University Press, 287-314.
Burkhardt, R. W. (1977). The Spirit of System: Lamarck and Evolutionary Biology.
Cambridge: Cambridge University Press.
Cantor, Geoffrey. (1983). Optics after Newton: Theories of Light in Britain and Ireland,
1704-1840. Manchester: Manchester University Press.
, and Hodge, M. J. S. (eds) (1981). Conceptions of Ether: Studies in the History
of Ether Theories. Cambridge: Cambridge University Press.
Carlson, E. A. (1966). The Gene: A Critical History. Philadelphia: Saunders.
Carnap, Rudolf. (1951). Logical Foundations of Probability. Chicago: University of
Chicago Press.
Carroll, Lewis. (1896). "What the Tortoise Said to Achilles," Mind, 4, 278-280.
Cartwright, Nancy. (1983). How the Laws of Physics Lie. Oxford: Oxford University
Press.
Cavendish, Henry. (1783/1961). The Composition of Water. Manchester: Alembic Club
Reprint.
Cherniak, Christopher. (1986). Minimal Rationality. Cambridge, MA: MIT Press.
Chisholm, Roderick. (1966). Theory of Knowledge. Englewood Cliffs, NJ: Prentice-
Hall.
Churchland, Patricia S. (1986). Neurophilosophy. Cambridge, MA: MIT Press.
Churchland, Paul M. (1979). Scientific Realism and the Plasticity of Mind. Cambridge:
Cambridge University Press.
. (1988). "Perceptual Plasticity and Theoretical Neutrality: A Reply to Jerry
Fodor," Philosophy of Science, 55, 167-187.
. (1989). A Neurocomputational Perspective. Cambridge, MA: MIT Press.
, and Hooker, Cliff, (eds) (1982). Images of Science. Chicago: University of
Chicago Press.
Cole, Jonathan, and Cole, Stephen. (1973). Social Stratification in Science. Chicago:
University of Chicago Press.
Coleman, William. (1964). Georges Cuvier. Zoologist. Cambridge, MA: Harvard
University Press.
Collins, H. M. (1985). Changing Order. London: Sage.
. (1987). "Pumps, Rock and Reality," Sociological Review, 21, 819-828.
394 Bibliography
, and Pinch, Trevor. (1982). Frames of Meaning. London: Routledge.
Conant, James Bryant. (1957). Harvard Case Histories in the Experimental Sciences.
Cambridge, MA: Harvard University Press.
Cosmides, Leda, and Tooby, John. (1988). "From Evolution to Behavior: Psychology
as the Missing Link," in (Dupre 1988), 277-306.
Crow, James. (1979). "Genes That Violate Mendel's Rules," Scientific American, 240,
no. 2, 134-146.
Culp, Sylvia, and Kitcher, Philip. (1989). "Theory Structure and Theory Change in
Contemporary Molecular Biology," British Journal for the Philosophy of Sci-
ence. 40, 459-483.
Cuvier, Georges. (1813). Essay on the Theory of the Earth. New York: Kirk and
Mercein.
Darwin, Charles. (1859). On The Origin of Species by Natural Selection, (first edition).
London: John Murray; facsimile reprint (1967), with introduction by Ernst
Mayr. Cambridge, MA: Harvard University Press.
. (1862). On the Various Contrivances by Which Orchids Are Fertilized by
Insects. London: John Murray.
. (1868). The Variation of Plants and Animals under Domestication. London:
John Murray.
Darwin, Francis. (1888). Life and Letters of Charles Darwin (three volumes). London:
John Murray; Reprint New York: Johnson Reprint, (1969).
. (1903). More Letters of Charles Darwin, (two volumes). London: John Murray.
Davidson, Donald. (1973). "On the Very Idea of a Conceptual Scheme," in Pro-
ceedings and Addresses of the American Philosophical Association, reprint
(1984) Essays on Truth and Interpretation. Oxford: Oxford University Press,
183-198.
Dawkins, Richard. (1976). The Selfish Gene. Oxford: Oxford University Press, (second
edition, with additional material) (1990). Oxford: Oxford University Press.
. (1982). The Extended Phenotype. San Francisco: Freeman.
. (1986). The Blind Watchmaker. London: Longmans.
De Candolle, Alphonse. (1855). Geographie Botanique Raisonee. Paris: Victor
Masson.
De Santillana, Georgio. (1955). The Crime of Galileo. Chicago: University of Chicago
Press.
Dennett, Daniel. (1988). The Intentional Stance. Cambridge, MA: MIT Press.
Devitt, Michael. (1981). Designation. New York: Columbia University Press.
. (1984). Realism and Truth. Princeton: Princeton University Press.
, and Sterelny, Kim. (1987). Language and Reality. Cambridge, MA: MIT Press.
Dobzhansky, Theodosius. (1937). Genetics and the Origin of Species (first edition);
(1982). reprint with introduction by Stephen Jay Gould. New York: Columbia
University Press.
Donnellan, Keith. (1974). "Speaking of Nothing," Philosophical Review, 83, 3-31.
Doppelt, Gerald. (1978). "Kuhn's Epistemological Relativism: An Interpretation and
Defense," Inquiry, 21, 33-86.
. (1986). "Relativism and the Reticulational Model of Scientific Rationality,"
Synthese, 69, 225-252.
. (1988). "The Philosophical Requirements for an Adequate Conception of
Scientific Rationality," Philosophy of Science, 55, 104-133.
Drake, Stillman. (1978). Galileo at Work. Chicago: University of Chicago Press.
Duhem, Pierre. (1906). The Aim and Structure of Physical Theory, (1951) translation
by Philip P. Wiener, Princeton: Princeton University Press.
Bibliography 395
Dummett, Michael. (1977). Elements of Intuitionism. Oxford: Oxford University
Press.
Dupre, John. (1981). "Natural Kinds and Biological Taxa," Philosophical Review,
90, 66-90.
. (1988). The Latest on the Best: Essays on Optimality and Evolution. Cam-
bridge, MA: MIT Press.
Earman, John. (1986). Determinism: A Primer. Dordrecht: Kluwer.
. (1992). Bayes, or Bust. Cambridge, MA: MIT Press.
, and Fine, Arthur. (1977). "Against Indeterminacy," Journal of Philosophy,
74, 535-538.
Eldredge, Niles. (1985). Unfinished Synthesis. New York: Oxford University Press.
, and Cracraft, Joel. (1980). Phylogenetic Patterns and the Evolutionary Process.
New York: Columbia University Press.
, and Gould, Stephen Jay. (1972). "Punctuated Equilibria: An Alternative to
Phyletic Gradualism," in T. J. Schopf (ed), Models in Paleobiology. San Fran-
cisco: Freeman.
Endler, John. (1986). Natural Selection in the Wild. Princeton: Princeton University
Press.
Evans, Gareth. (1975). "The Causal Theory of Names," Proceedings of the Aristotelian
Society, 47, 187-208.
Faust, David. (1985). The Limits of Scientific Reasoning. Minneapolis: University of
Minnesota Press.
Feigl, Herbert. (1950). "De Principiis Non Est Disputandum. . . ?" in Max Black
(ed), Philosophical Analysis (1963), reprint. Englewood Cliffs, NJ: Prentice-
Hall, 113-131.
Feyerabend, Paul. (1963a). "Explanation, Reduction, and Empiricism," in Minnesota
Studies in the Philosophy of Science, volume III. Minneapolis: University of
Minnesota, 28-97.
. (1963b). "How to Be a Good Empiricist," Proceedings of the Delaware Sem-
inar, volume 2. New York: Interscience, 3-39.
. (1965). "Problems of Empiricism," in R. Colodny (ed), Beyond the Edge of
Certainty. Englewood Cliffs, NJ: Prentice-Hall, 145-260.
. (1970). "Against Method," in Minnesota Studies in the Philosophy of Science,
volume IV. Minneapolis: University of Minnesota, 17-130.
. (1975). Against Method. London: Verso.
. (1979). Science in a Free Society. London: New Left Books.
. (1987). Farewell to Reason. London: Verso.
Feynman, Richard et al. (1963). The Feynman Lectures on Physics. Reading, MA:
Addison-Wesley.
Field, Hartry. (1972). "Tarski's Theory of Truth," Journal of Philosophy, 69, 347-
375.
. (1973). "Theory Change and the Indeterminacy of Reference," Journal of
Philosophy, 70, 462-481.
. (1978). "A Note on Jeffrey Conditionalization," Philosophy of Science. 45,
361-367.
Fine, Arthur. (1986). The Shaky Game. Chicago: University of Chicago Press.
Finocchiaro, Maurice. (1989). The Galileo Affair. Berkeley: University of California
Press.
Fisher, R. A. (1918). "The Correlation between Relatives on the Supposition of
Mendelian Inheritance," Transactions of the Royal Society of Edinburgh, 52,
399-433.
396 Bibliography
. (1930). The Genetical Theory of Natural Selection. Oxford: Oxford University
Press. (1956) reprint New York: Dover.
Fodor, Jerry. (1984). "Observation Reconsidered," Philosophy of Science, 51, 23-43.
Foley, Richard. (1988a). Epistemic Rationality. Cambridge, MA: Harvard University
Press.
. (1988b). "Some Different Conceptions of Rationality," in E. McMullin (ed),
Construction and Constraint. Notre Dame: University of Notre Dame Press,
123-152.
Frege, Gottlob. (1892). "On Sense and Reference," in P. Geach and M. Black (eds),
(1952) Translations from the Writings of Gottlob Frege. Oxford, Blackwell, 56
78.
Fresnel, Augustin. (1818). "Memoire sur la Diffraction de la Lumiere (Coronn6 par
1'Academie des Sciences)," in (Fresnel 1865/1965), volume I, 247-382.
. (1865/1965). Oeuvres. (three volumes). Paris: Imprimerie Imperiale.
Friedman, Michael. (1974). "Explanation and Scientific Understanding," Journal of
Philosophy, 71, 5-19.
Fuller, Steven. (1988). Social Epistemology. Bloomington: Indiana University Press.
Galileo Galilei. Dialogue Concerning the Two Great Systems of the World. Berkeley:
University of California Press.
. The Starry Messenger (trans Albert van Helden). Chicago: University of Chi-
cago Press.
Galileo/Favaro. (1890-1909). [Galileo Galilei, edited by Antonio Favaro] Opere. Flor-
ence: Barbera.
Galison, Peter. (1987). How Experiments End. Chicago: University of Chicago Press.
Garber, Daniel. (1982). "Old Evidence and Logical Omniscience," in Minnesota
Studies in the Philosophy of Science, volume X. Minneapolis: University of
Minnesota, 99-131.
Ghiselin, Michael. (1969). The Triumph of the Darwinian Method. Berkeley: Uni-
versity of California Press.
. (1974). "A Radical Solution to the Species Problem," Systematic Zoology.
23, 536-544.
. (1989). Intellectual Compromise: The Bottom Line. New York: Paragon.
Giere, Ronald. (1985). "Philosophy of Science Naturalized," Philosophy of Science.
52, 331-356.
. (1988). Explaining Science. Chicago: University of Chicago Press.
Gingerich, Owen, and Westman, Robert. (1991). "The Wittich Connection" Trans-
actions of the American Philosophical Society. 78, part 7.
Glick, Thomas, (ed) (1972). The Comparative Reception of Darwinism. Austin: Uni-
versity of Texas Press.
Glymour, Clark. (1980). Theory and Evidence. Princeton: Princeton University Press.
Goldman, Alvin. (1979). "What Is Justified Belief?" in G. Pappas (ed), Justification
and Knowledge. Dordrecht: Reidel, 1-23.
. (1986). Epistemology and Cognition. Cambridge, MA: Harvard University
Press.
. (1987). "Foundations of Social Epistemics," Synthese, 73, 109-144.
. (1989). "Epistemic Folkways and Justification and Scientific Epistemology,"
in A. Goldman (ed), Liaisons. Cambridge, MA: MIT Press, 1992, 155-175.
Goodman, Nelson. (1955). Fact, Fiction, and Forecast. Indianapolis: Bobbs-Merrill.
Gould, Stephen Jay. (1977). Ontogeny and Phylogeny. Cambridge, MA: Harvard
University Press.
Bibliography 397
. (1977a). Ever Since Darwin. New York: Norton.
. (1980). The Panda's Thumb. New York: Norton.
. (1980a). "The Return of the Hopeful Monster," in (Gould 1980), 186-193.
. (1980b). "G. G. Simpson, Paleontology, and the Modern Synthesis," in (Mayr
and Provine 1980).
. (1980c). "Is a New and General Theory of Evolution Emerging?" Paleo-
biology, 6, 119-130.
. (1981). The Mismeasure of Man. New York: Norton.
. (1982). "Darwinism, and the Expansion of Evolutionary Theory," Science,
216, 380-387.
. (1983). "The Hardening of the Modern Synthesis," in M. Grene, Dimensions
of Darwinism. Cambridge: Cambridge University Press, 71-93.
. (1989). Wonderful Life. New York: Norton.
, and Lewontin, Richard C. (1979). "The Spandrels of San Marco and the
Panglossian Paradigm: A Critique of the Adaptationist Programme," Proceed-
ings of the Royal Society, B 205, 581-598; reprint in E. Sober (ed) (1984).
Conceptual Issues in Evolutionary Biology. Cambridge, MA: MIT Press, 252-
270.
Grandy, Richard. (1973). "Reference, Meaning, and Belief," Journal of Philosophy,
70, 439-452.
Gray, Asa. (1876). Darwiniana. New York: Appleton.
Hacking, Ian. (1983). Representing and Intervening. Cambridge: Cambridge University
Press.
Hagstrom, Warren. (1965). The Scientific Community. New York: Basic Books.
Hall, Mary Boas. (1966). Robert Boyle on Natural Science. Bloomington: Indiana
University Press.
Hamilton, W. D. (1966). "The Genetical Evolution of Social Behavior," (parts I and
II), Journal of Theoretical Biology, 7, 1-52.
Hanna, Joseph. (1985). "Sociobiology and the Information Metaphor," in James
Fetzer (ed), Sociobiology and Explanation. Dordrecht: Reidel, 31-56.
Hanson, Norwood Russell. (1958). Patterns of Discovery. Cambridge: Cambridge
University Press.
Hardwig, John. (1985). "Epistemic Dependence," Journal of Philosophy, 82, 335-
49.
Hardy, G.H. (1940). A Mathematician's Apology. Oxford: Oxford University Press.
Harman, Gilbert. (1971). Thought. Princeton: Princeton University Press.
. (1987). Change in View. Cambridge, MA: MIT Press.
Haugeland, John. (1985). Artificial Intelligence: The Very Idea. Cambridge, MA: MIT
Press.
Heinrich, Bernd. (1981). Bumblebee Economics. Cambridge, MA: Harvard University
Press.
Hempel, Carl G. (1945). "Studies in the Logic of Confirmation," in (Hempel 1965),
3-46.
. (1950). "Problems and Changes in the Empiricist Criterion of Meaning,"
Revue Internationale de Philosophie, 11, 41-63.
. (1951). "The Concept of Cognitive Significance: A Reconsideration," Pro-
ceedings of the American Academy of Arts and Sciences, 80, no. 1, 61-77.
. (1965). Aspects of Scientific Explanation. New York: Free Press.
. (1966). Philosophy of Natural Science. Englewood Cliffs, NJ: Prentice-Hall.
Hesse, Mary. (1962). Forces and Fields. Westport, CT: Greenwood.
398 Bibliography
. (1970). "Is There an Independent Observation-Language?" in R. Colodny
(ed), The Nature of Scientific Theories. Pittsburgh: University of Pittsburgh
Press, 35-77.
Ho, Mae-Wan, and Fox, Sydney. (1987). Metaphors in The New Evolutionary Par-
adigm. New York: Wiley.
Ho, Mae-Wan, and Saunders, Peter. (1984). Beyond Neo-Darwinism: An Introduction
to the New Evolutionary Paradigm. New York: Academic Press.
Hodge, M. J. S. (1971). "Species in Lamarck," in J. Schuller (ed), Colloque Inter-
nationale "Lamarck." Paris: A Blanchard, 31-46.
. (1977). "The Structure and Strategy of Darwin's 'Long Argument'," British
Journal for the History of Science, 10, 237-246.
Holland, John; Holyoak, Keith; Nisbett, Richard; and Thagard, Paul. (1986). Induc-
tion. Cambridge, MA: MIT Press.
Holmes, Frederic L. (1985). Lavoisier and the Chemistry of Life. Madison: University
of Wisconsin.
. (1988). "Lavoisier's Conceptual Passage," Osiris, 4, 82-92.
Horwich, Paul. (1982). Probability and Evidence. Cambridge: Cambridge University
Press.
Hosiasson-Lindenbaum, Janina. (1940). "On Confirmation," Journal of Symbolic
Logic, 5, 133-148.
Howson, Colin, and Urbach, Peter. (1989). Scientific Reasoning. LaSalle, IL: Open
Court.
Hull, David. (1973a). Philosophy of Biological Science. Englewood Cliffs, NJ: Pren-
tice-Hall.
. (ed) (1973b). Darwin and His Critics. Cambridge, MA: Harvard University
Press.
. (1978). "A Matter of Individuality," Philosophy of Science. 45, 335-360.
. (1982). "Darwinism as a Historical Entity," in P. Asquith and T. Nickles
(eds), PSA 1982. East Lansing, MI: Philosophy of Science Association.
. (1988). Science as a Process. Chicago: University of Chicago Press.
Huxley, Leonard. (1913). The Life and Letters of Thomas Henry Huxley. New York:
Appleton.
Huxley, Thomas Henry. (1896). Darwiniana. New York: Appleton.
Hyde, T. S., and Jenkins J. J. (1973). "Recall for Words as a Function of Semantic,
Graphic, and Syntactic Orienting Tasks," Journal of Verbal Learning and Ver-
bal Behavior, 12, 471-480.
Jeffrey, Richard. (1983). The Logic of Decision (second edition). Chicago: University
of Chicago Press.
Kant, Immanuel. (1781/1968). Critique of Pure Reason, trans. Norman Kemp Smith.
London: MacMillan.
Kim, Jaegwon. (1988). "What Is Naturalistic Epistemology?," Philosophical Per-
spectives, 2, 381-405.
Kirwan, Richard. (1783). An Essay on Phlogiston (first edition). London.
. (1789). An Essay on Phlogiston; reprint (Kirwan 1782) with translations of
comments by Lavoisier and other French chemists, which had been included
in a French translation of (Kirwan 1782), and replies by Kirwan. London: J.
Johnson (both editions).
Kitcher, Patricia (1985). "Narrow Taxonomy and Wide Functionalism," Philosophy
of Science, 52, 78-97.
Bibliography 399
. (1988). "Marr's Computational Theory of Vision," Philosophy of Science, 55,
1-24.
Kitcher, Patricia. (1990). Kant's Transcendental Psychology. New York: Oxford Uni-
versity Press.
Kitcher, Philip. (1978). "Theories, Theorists, and Theoretical Change," Philosophical
Review. 87, 519-547.
. (1979). "Frege's Epistemology," Philosophical Review, 88, 235-262.
. (1981). "Explanatory Unification," Philosophy of Science, 48, 507-531.
. (1982). "Genes," British Journal for the Philosophy of Science, 33, 337-359.
. (1982b). Abusing Science: The Case against Creationism. Cambridge, MA:
MIT Press.
. (1983a). The Nature of Mathematical Knowledge. New York: Oxford Uni-
versity Press.
. (1983b). "Kant's Philosophy of Science," Midwest Studies in Philosophy, 8,
387-408.
. (1983c). "Implications of Incommensurability," in P. Asquith and T. Nickles
(eds), PSA 1982, volume II. East Lansing, MI: PSA, 689-703.
. (1984). "1953 and All That: A Tale of Two Sciences," Philosophical Review,
93, 335-373.
. (1985a). "Darwin's Achievement," in Nicholas Rescher (ed), Reason and
Rationality in Science. Washington, DC: University Press of America, 123-185.
. (1985b). Vaulting Ambition: Sociobiology and the Quest for Human Nature.
Cambridge, MA: MIT Press.
. (1986). "Projecting the Order of Nature," in Robert Butts (ed), Kant's Phi-
losophy of Physical Science. Dordrecht: Reidel, 201-235.
. (1988). "Why Not the Best?," in (Dupre 1988), 77-102.
. (1989). "Explanatory Unification and the Causal Structure of the World," in
P. Kitcher and W. Salmon (eds), Scientific Explanation. Minneapolis: Univer-
sity of Minnesota, 410-505.
. (1990). "Developmental Decomposition and the Future of Human Behavioral
Ecology," Philosophy of Science, 57, 96-117.
. (1992). "The Naturalists Return," Philosophical Review, 101, 53-114.
, and Salmon, Wesley, (eds) (1989). Scientific Explanation. Minneapolis: Uni-
versity of Minnesota.
Kohn, David. (ed) (1985). The Darwinian Heritage. Princeton: Princeton University
Press.
Kordig, Carl. (1971). The Justification of Scientific Change. Dordrecht: Reidel.
Kornblith, Hilary. (1980). "Beyond Foundationalism and the Coherence Theory,"
Journal of Philosophy, 72, 597-612.
. (1985). "Editorial Introduction" in Naturalistic Epistemology. Cambridge,
MA: MIT Press.
Kosslyn, Stephen. (1981). Image and Mind. Cambridge, MA: Harvard University
Press.
Kripke, Saul. (1972). "Naming and Necessity," in D. Davidson and G. Harman (eds),
Semantics of Natural Language. Dordrecht: Reidel; reprinted as (Kripke 1980),
253-355.
. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.
Kuhn, Thomas S. (1962). The Structure of Scientific Revolutions, (first edition). Chi-
cago: University of Chicago Press.
400 Bibliography
. (1962/70). The Structure of Scientific Revolutions. (first or second). Chicago:
University of Chicago Press.
. (1970). The Structure of Scientific Revolutions. (second edition). Chicago:
University of Chicago Press.
. (1977). The Essential Tension. Chicago: University of Chicago Press.
. (1983). "Commensurability, Comparability, Communicability," in P. Asquith
and T. Nickles (eds), PSA 1982. East Lansing, MI: PSA, 669-688.
Lakatos, Imre. (1970). "Falsification and the Methodology of Scientific Research
Programmes," in I. Lakatos and A. Musgrave (eds), Criticism and the Growth
of Knowledge. Cambridge: Cambridge University Press, 91-196.
. (1971). "History of Science and its Rational Reconstructions," in I. Lakatos
(ed), Philosophical Papers, volume I. Cambridge: Cambridge University Press,
102-139.
, and Zahar, Elie. (1976). "Why Did Copernicus' Research Programme Succeed
Ptolemy's?," in (Westman 1976), 354-383.
Lande, R. (1981). "Models of speciation by sexual selection on polygenic traits,"
Proceedings of the National Academy of Sciences, 78, 3721-3725.
Latour, Bruno. (1987). Science in Action. Cambridge, MA: Harvard University Press.
. (1988). The Pasteurization of France. Cambridge, MA: Harvard University
Press.
. (1989a). "Postmodern? No, Simply AModern : Steps towards an Anthropology
of Science," Studies in the History and Philosophy of Science, 21, 145-171.
. (1992). "One More Turn after the Social Turn . . .," in E. McMullin, The
Social Dimensions of Scientific Knowledge. Notre Dame: University of Notre
Dame Press, 272-294.
, and Woolgar, Steve. (1979). Laboratory Life. London: Sage, (1986) reprint
Princeton: Princeton University Press.
Laudan, Larry. (1977). Progress and Its Problems. Berkeley: University of California
Press.
. (1981). "A Confutation of Convergent Realism," Philosophy of Science, 48.
. (1984). Science and Values. Berkeley: University of California Press.
. (1987). "Progress or Rationality: The Prospects for Normative Naturalism,"
American Philosophical Quarterly, 24, 19-31.
. (1991). Relativism: A Dialogue. Chicago: University of Chicago Press.
Lavoisier, Antoine-Laurent. (1774). "Analyse du Memoire sur 1'Augmentation du
Poids des Metaux par la Calcination," in (Lavoisier 1862), II, 97-99.
. (1777). "Memoire sur la Combustion en General," in (Lavoisier 1862), II,
225-233.
. (1782). "Memoire sur 1'Union du Principe Oxygine avec le Per," in (Lavoisier
1862), II, 557-574.
. (1783). "Reflexions sur le Phlogistique," in (Lavoisier 1862), II, 623-655.
. (1789). Traite Elementaire de Chimie. Paris: Cuchet.
. (1792). "Details Historiques sur la Cause de 1'Augmentation de Poids
qu'acquierent les substances Metalliques, lorsqu'on les chauffe pendant leur
exposition a l'air," in (Lavoisier 1862), II, 99-103.
. (1862). Oeuvres. Paris: Imprimerie Imperiale.
Leplin, Jarrett. (ed) (1982). Scientific Realism. Berkeley: University of California
Press.
Levi, Isaac. (1982). The Enterprise of Knowledge. Cambridge, MA: MIT Press.
Bibliography 401
Levins, Richard. (1966). Evolution in Changing Environments. Princeton: Princeton
University Press.
, and Lewontin, Richard C. (1985). The Dialectical Biologist. Cambridge, MA:
Harvard University Press.
Lewis, David. (1973a). Counterfactuals. Oxford: Blackwell.
. (1973b). "Causation," Journal of Philosophy, 70, 556-567.
Lewontin, Richard C. (1968). " 'Honest Jim' Watson's Big Thriller about DNA," in
Gunnar Stent (ed), The Double Helix. New York: Norton, 185-187.
. (1970). "The Units of Selection," Annual Review of Ecology and Systematics,
1, 1-18.
. (1974). The Genetic Basis of Evolutionary Change. New York: Columbia
University Press.
. (1978). "Adaptation," Scientific American, 239, no. 3, 212-230.
. Rose, Stephen, and Kamin, Leon. (1984). Not in Our Genes. New York:
Pantheon.
Lloyd, Elisabeth. (1983). "The Nature of Darwin's Support for the Theory of Natural
Selection," Philosophy of Science, 50, 112-129.
. (1984). "A Semantic Approach to the Structure of Population Genetics,"
Philosophy of Science, 51, 242-264.
. (1988). The Structure and Confirmation of Evolutionary Theory. Westport,
CT: Greenwood.
Loftus, Elizabeth. (1979). Eyewitness Testimony. Cambridge, MA: Harvard University
Press.
Longino, Helen. (1990). Science as Social Knowledge. Princeton: Princeton University
Press.
MacArthur, Robert, and Wilson, E. O. (1967). The Theory of Island Biogeography.
Princeton: Princeton University Press.
Machamer, Peter. (1973). "Feyerabend and Galileo: The Interaction of Theories and
the Reinterpretation of Experience," Studies in the History and Philosophy of
Science, 4, 1-46.
Mackie, J. L. (1963). "The Paradox of Confirmation," British Journal for the Phi-
losophy of Science, 13, 265-277.
. (1973). The Cement of the Universe. Oxford: Oxford University
Press.
Manier, Edward. (1978). The Young Darwin and His Cultural Circle. Dordrecht:
Reidel.
Margolis, Howard. (1982). Selfishness, Altruism, and Rationality: A Theory of Social
Choice. Cambridge: Cambridge University Press.
. (1987). Patterns, Judgment, and Cognition. Chicago: University of Chicago
Press.
Marr, David. (1982). Vision. San Francisco: Freeman.
Martin, Paul, and Bateson, Patrick. (1986). Measuring Behavior: An Introductory
Guide. Cambridge: Cambridge University Press.
Maxwell, Grover. (1963). "The Ontological Status of Theoretical Entities," in Min-
nesota Studies in the Philosophy of Science, volume III. Minneapolis: University
of Minnesota, 3-37.
Maynard Smith, John. (1982). Evolution and the Theory of Games. Cambridge: Cam-
bridge University Press.
. (1988). "How to Model Evolution," in (Dupre 1988), 119-132.
402 Bibliography
. (1989). Did Darwin Get It Right?. London: Chapman and Hall.
Mayr, Ernst. (1942). Systematics and the Origin of Species. New York: Columbia
University Press.
. (1963). Animal Species and Evolution. Cambridge, MA: Harvard University
Press.
. (1976). Evolution and the Diversity of Life. Cambridge, MA: Harvard Uni-
versity Press.
. (1982). The Growth of Biological Thought. Cambridge, MA: Harvard Uni-
versity Press.
, and Provine, William. (eds) (1980). The Evolutionary Synthesis. Cambridge,
MA: Harvard University Press.
McClelland, Jay, and Rumelhart, David. (1986). Parallel Distributed Processing. Cam-
bridge, MA: MIT Press.
McEvoy, John. (1988). "Continuity and Discontinuity in the Chemical Revolution,"
Osiris, 4, 195-213.
McMullin, Ernan. (ed) (1988). Construction and Constraint. Notre Dame: University
of Notre Dame Press.
Merton, Robert K. (1973). The Sociology of Science. Chicago: University of Chicago
Press.
Mill, John Stuart. (1859). On Liberty. Indianapolis: Bobbs-Merrill.
Miller, David. (1974). "On the Comparison of False Theories by Their Bases," British
Journal for the Philosophy of Science, 25, 166-177.
. (1979). "The Accuracy of Predictions," Synthese, 30, 159-191.
Miller, George. (1956). "The Magic Number Seven Plus or Minus Two," Psychological
Review, 63, 81-97.
Milne-Edwards, Henri. (1844). "Considerations sur quelques principes relatifs a la
classification naturelle des animaux," Annales des Sciences Naturelles, 3rd Se-
ries, 1, 65-99.
Mondadori, Fabrizio, and Morton, Adam. (1976). "Modal Realism: The Poisoned
Pawn," Philosophical Review, 85 3-20.
Moore, James. (1985). "Darwin of Down: The Evolutionist as Squarson-Naturalist,"
in (Kohn 1985), 435-482.
Newton-Smith, William. (1981). The Rationality of Science. London: Routledge.
Nisbett, Richard, and Ross, Lee. (1980). Human Inference: Strategies and Shortcom-
ings of Social Judgment. Englewood Cliffs, NJ: Prentice-Hall.
Olby, R. C. (1974). The Path to the Double Helix. London: Macmillan.
Ospovat, Dov. (1981). The Development of Darwin's Theory: Natural History, Natural
Theology, and Natural Selection 1838-1859. Cambridge: Cambridge University
Press.
Owen, Richard. (1848). On the Archetype and Homologies of the Vertebrate Skeleton.
London: Van Voorst.
. (1849). On the Nature of Limbs. London: Van Voorst.
Papineau, David. (1988). Reality and Representation. Oxford: Blackwell.
Partington, J. R. (1965). A Short History of Chemistry. London: MacMillan.
, and McKie, Douglas. (1937-38). "Historical Studies on the Phlogiston The-
ory," Annals of Science, 2, 361-404, 4, 1-58, 4, 337-371.
Peacocke, Christopher. (1987). Thoughts: An Essay on Content. Oxford: Blackwell.
Peirce, C. S. (1958). Collected Works. Cambridge, MA: Harvard University Press.
Perrin, Carl B. (1988). "Research Traditions, Lavoisier, and the Chemical Revolu-
tion," Osiris, 4, 53-81.
Bibliography 403
Pickering, Andrew. (1984). Constructing Quarks. Chicago: University of Chicago
Press.
Pinch, Trevor. (1985). "Theory Testing in Sciencethe Case of the Solar Neutrinos,"
Philosophy of Social Science, 15, 167-188.
. (1986). "Strata Various," Social Studies of Science, 16, 705-713.
Popper, Karl R. (1959). The Logic of Scientific Discovery. London: Hutchinson.
Powell, Baden. (1855). Essays on the Spirit of the Inductive Philosophy, the Unity of
Worlds, and the Philosophy of Creation. London: Longmans.
Priestley, Joseph (1775/1970). Experiments and Observations on Different Kinds of
Air. New York: Johnson Reprint.
Provine, William. (1971). The Origins of Theoretical Population Genetics. Chicago:
University of Chicago Press.
. (1986). Sewall Wright and Evolutionary Biology. Chicago: University of Chi-
cago Press.
Putnam, Hilary. (1973). "Meaning and Reference," Journal of Philosophy, 70, 699-
711.
. (1978). Meaning and the Moral Sciences. London: Routledge.
. (1981). Reason, Truth, and History. Cambridge: Cambridge University Press.
Quine, W. V. (1951). "Two Dogmas of Empiricism," in From a Logical Point of
View. New York: Harper, 20-46.
. (1960). Word and Object. Cambridge, MA MIT Press.
. (1966). "Carnap and Logical Truth," in The Ways of Paradox. New York:
Random House, 107-132.
. (1970). Ontological Relativity and Other Essays. New York: Columbia Uni-
versity Press.
. (1970a). "Epistemology Naturalized," in (Quine 1970), 69-90.
. (1970b). "Natural Kinds" in (Quine 1970), 114-138.
. (1974). The Roots of Reference. LaSalle, IL: Open Court.
Redondi, Pietro. (1987). Galileo, Heretic. Princeton: Princeton University Press.
Rescher, Nicholas. (1989a). A Useful Inheritance. Totowa, NJ: Rowman and
Littlefield.
. (1989b). Cognitive Economy. Pittsburgh: University of Pittsburgh Press.
Rosenberg, Alexander. (1978). "The Supervenience of Biological Concepts," Phi-
losophy of Science, 45, 368-386.
. (1983). "Fitness," Journal of Philosophy, 80, 457-473.
. (1985). The Structure of Biological Science. New York: Cambridge University
Press.
Rosenkrantz, R. D. (1977). Inference, Method and Decision: Towards a Bayesian
Philosophy of Science. Dordrecht: Reidel.
Roughgarden, Jonathan. (1979). Theory of Population Genetics and Evolutionary
Ecology: An Introduction. New York: Macmillan.
Rouse, Joseph. (1987). Knowledge and Power. Ithaca: Cornell University Press.
Rudwick, Martin J. S. (1972/85). The Meaning of Fossils. Chicago: University of
Chicago; (1985) reprint.
. (1970). "The Strategy of Lyell's Principles of Geology," Isis, 61, 5-33.
. (1985). The Great Devonian Controversy. Chicago: University of Chicago
Press.
Ruse, Michael. (1973). Philosophy of Biology. London: Hutchinson.
. (1979). The Darwinian Revolution. Chicago: University of Chicago Press.
. (1986). Taking Darwin Seriously. Oxford: Blackwell.
404 Bibliography
Russell, E. B. (1916). Form and Function. Chicago: University of Chicago Press.
Salmon, Wesley. (1957). "Should We Attempt to Justify Induction?," Philosophical
Studies, 8, 33-48.
. (1967). Foundations of Scientific Inference. Pittsburgh: University of Pittsburgh
Press.
. (1968). "The Justification of Inductive Rules of Inference," in I. Lakatos (ed),
The Problem of Inductive Logic. Amsterdam: North-Holland, 24-43.
. (1970). "Bayes' Theorem and the History of Science," Minnesota Studies in
the Philosophy of Science, volume V. Minneapolis: University of Minnesota
Press, 68-86.
. (1984). Scientific Explanation and the Causal Structure of the World. Princeton:
Princeton University Press.
. (1989). Four Decades of Scientific Explanation. Minneapolis: University of
Minnesota; also in (Kitcher and Salmon 1989).
. (1990). "Rationality and Objectivity in Science or Tom Kuhn Meets Tom
Bayes," Minnesota Studies in the Philosophy of Science, volume XIV. Min-
neapolis: University of Minnesota Press, 175-204.
Sarkar, Husain. (1982). A Theory of Method. Berkeley: University of California Press.
Schaffner, Kenneth. (1972). Nineteenth-Century Ether Theories. Oxford: Pergamon.
Scheffler, Israel. (1967). Science and Subjectivity. Indianapolis: Bobbs-Merrill.
Schelling, Thomas. (1984). Choice and Consequence. Cambridge, MA: Harvard Uni-
versity Press.
Schlick, Moritz (1934/1959). "The Foundation of Knowledge," in A. J. Ayer (ed),
Logical Positivism. New York: Free Press, 209-227.
Schofield, Robert. (1967). A Scientific Autobiography of Joseph Priestley. Cambridge,
MA: MIT Press.
. (1969). Mechanism and Materialism. Princeton: Princeton University Press.
Secord, James. (1985). "Darwin and the Breeders," in (Kohn 1985), 519-542.
Sellars, Wilfrid. (1956). "Empiricism and the Philosophy of Mind," in (Sellars 1963),
chapter 5.
. (1963). Science, Perception, and Reality. London: Routledge.
. (1967). Science and Metaphysics. London: Routledge.
Shapere, Dudley. (1982). "The Concept of Observation in Science and in Philosophy,"
Philosophy of Science, 49, 485-525.
. (1984). Reason and the Growth of Knowledge. Dordrecht: Reidel.
Shapin, Steven. (1982). "History of Science and Its Sociological Reconstructions,"
History of Science. 20, 157-211.
, and Schaffer, Simon. (1985). Leviathan and the Air-Pump. Princeton: Prince-
ton University Press.
Shepard, R., and Chipman, S. (1970). "Second-order Isomorphism of Internal Rep-
resentations: Shapes of States," Cognitive Psychology, 1, 1-17.
, and Metzler, J. (1971). "Mental Rotation of Three-Dimensional Objects,"
Science. 171, 701-703.
Simon, Herbert. (1957). Models of Man. New York: Wiley.
Simpson, George Gaylord. (1944). Tempo and Mode in Evolution. New York: Co-
lumbia University Press.
. (1953). The Major Features of Evolution. New York: Columbia University
Press.
Skyrms, Brian. (1986). Choice and Chance (third edition). Belmont, CA:
Wadsworth.
Bibliography 405
Sneed, Joseph. (1971). The Logical Structure of Mathematical Physics. Dordrecht:
Reidel.
Sober, Elliott. (1984). The Nature of Selection. Cambridge, MA: MIT Press.
. (1988). Review of (Salmon 1984), British Journal for the Philosophy of Science,
38, 243-257.
. (1988). Reconstructing the Past. Cambridge, MA: MIT Press.
Sokal, Robert, and Rohlf, James. (1974). Biometry. San Francisco: Freeman.
Solomon, Miriam. (1992). "Scientific Rationality and Human Reasoning," Philosophy
of Science (forthcoming).
Stanley, Steven. (1979). Macroevolution, Pattern and Process. San Francisco:
Freeman.
Sterelny, Kim, and Kitcher, Philip. (1988). "The Return of the Gene," Journal of
Philosophy, 85, 339-361.
Stich, Stephen. (1983). The Case against Belief: From Folk Psychology to Cognitive
Science. Cambridge, MA: MIT Press.
. (1985). "Could Man Be an Irrational Animal?," Synthese, 64, 115-134.
. (1990). The Fragmentation of Reason. Cambridge, MA: MIT Press.
Strawson, Peter. (1952). Introduction to Logical Theory. London: Methuen.
Strum, Shirley. (1987). Almost Human. New York: Random House.
Sulloway, Frank. (1982). "Darwin's Finches, the Evolution of a Legend," Journal for
the History of Biology, 15, 1-53.
. (forthcoming). "Orthodoxy and Innovation in Science," manuscript.
Suppe, Fred. (1972). "What's Wrong with the Received View of Scientific Theories?,"
Philosophy of Science, 39, 1-19.
. (ed) (1977). The Structure of Scientific Theories (second edition). Urbana:
University of Illinois Press.
Suppes, Patrick. (1967). "What Is a Scientific Theory?" in A. Danto and S. Morgen-
besser (eds), Philosophy of Science Today. New York: Basic Books, 55-67.
Tarski, Alfred. (1936). "The Concept of Truth in Formalized Languages," in J. H.
Woodger(ed. and trans.), Logic, Semantics, Metamathematics. Oxford: Oxford
University Press (1956), 152-278.
. (1944). "The Semantic Conception of Truth," in H. Feigl and W. Sellars (eds),
Readings in Philosophical Analysis. New York: Appleton (1949), 341-374.
Toulmin, Stephen. (1961). Foresight and Understanding. New York: Harper and Row.
. (1972). Human Understanding. Princeton: Princeton University Press.
Tversky, Amos, and Kahneman, Daniel. (1973). "Availability: A Heuristic for Judg-
ging Frequency and Probability," Cognitive Psychology, 5, 207-232.
. (1974). "Judgment under Uncertainty: Heuristics and Biases," Science. 185,
1124-1131.
Van Fraassen, Bas. (1980). The Scientific Image. Oxford: Oxford University Press.
. (1985). "Salmon on Explanation," Journal of Philosophy, 82, 639-651.
Van Helden, Albert. (1977). The Invention of the Telescope. Philadelphia: American
Philosophical Society Monograph.
. (1989). Introduction to translation of Galileo's Sidereus Nuncius. Chicago:
University of Chicago Press.
Watson, James. (1967). The Double Helix. New York: Norton.
Westfall, R. S. (1982). Never at Rest. Cambridge: Cambridge University Press.
. (1985). "Science and Patronage: Galileo and the Telescope," Isis, 76, 11-30.
Westman, Robert. (1975). "The Melanchthon Circle, Rheticus, and the Wittenberg
Interpretation of Copernican Theory," Isis, 165-193.
406 Bibliography
. (ed) (1976). The Copernican Achievement. Los Angeles: University of Cali-
fornia Press.
. (1980). "The Astronomer's Role in the Sixteenth Century: A Preliminary
Study," History of Science, 18, 105-142.
. (1990). "Proof, Poetics, and Patronage: Copernicus' Preface to De Revolu-
tionibus," in Westman and David Lindberg (eds), Reappraisals of the Scientific
Revolution. Cambridge: Cambridge University Press, 167-205.
Williams, George C. (1966). Adaptation and Natural Selection. Princeton: Princeton
University Press.
Williams, Mary. (1970). "Deducing the Consequences of Evolution," Journal of The-
oretical Biology, 29, 343-385.
Wilson, David S. (1980). The Natural Selection of Populations and Communities.
Menlo Park: Benjamin/Cummings.
. (1983). "The Group Selection Controversy: History and Current Status,"
Annual Review of Ecology and Systematics, 14, 159-187.
Wilson, E. O.; Carpenter, F. M.; and Brown, W. L. (1967). "The First Mesozoic
Ants," Science, 157, 1038-1040.
Wittgenstein, Ludwig. (1953). Philosophical Investigations. Oxford: Blackwell.
Woodward, James. (1989). "The Causal-Mechanical Model of Explanation," in
(Kitcher and Salmon 1989), 357-383.
Woolfenden, G., and Fitzpatrick, J. (1984). The Florida Scrubjay. Princeton: Prince-
ton University Press.
Worrall, John. (1976). "Thomas Young and the 'Refutation' of Newtonian Optics: A
Case Study in the Interaction of Philosophy of Science and the History of
Science," in Colin Howson (ed), Method and Appraisal in the Physical Sciences.
Cambridge: Cambridge University Press, 107-180.
. (1978). "The Ways in Which the Methodology of Scientific Research Pro-
grammes Improves on Popper's Methodology," in G. Radnitzky and G. An-
derson (eds), Progress and Rationality in Science. Dordrecht: Reidel, 45-70.
. (1985). "Scientific Discovery and Theory-Confirmation," in J. Pitt (ed),
Change and Progress in Modern Science. Dordrecht: Reidel, 301-332.
. (1988). "The Value of a Fixed Methodology," British Journal for the Philos-
ophy of Science, 39, 263-275.
. (1989). "Fresnel, Poisson, and the White Spot," in D. Gooding et al. (eds),
The Uses of Experiment. Cambridge: Cambridge University Press, 135-157.
Wright, Sewall. (1931). "Evolution in Mendelian Populations," Genetics, 16, 97-159.
Index
innovation, community response to, 334, Kuhn, Thomas, 6, 7n, 9n, 64n, 67, 75, 84n,
339-44 87n, 89, 93n, 97-98, 103, 110, 113,
inquiry, scientific, 73, 115, 125 115-16n, 124, 127n, 128-32, 133n,
institutions, 9n, 73, 89, 318, 351 140n, 141n, 167n, 173-75, 177,
instrumentalism, 134 178n, 202n, 220, 223n, 225, 227n,
instruments, 60, 74-75, 85-86, 95, 117, 251-52, 260n, 262, 272-76, 294-95,
118n, 123-24, 130, 221, 228-34, 344n
259, 300, 308. See also practices, in-
dividual L-practices, I58df, 159
intensional entities, 75 labor, cognitive
intentions, 61n, 77n, 309 division/distribution of, 77n, 89, 304-5,
interactions, 60-61, 217n, 220-21n, 228, 310, 334, 337, 342, 344-55, 356n,
243, 263n, 323 357, 360, 362, 365, 374, 376, 381,
interference, social, 160-69 388
intrinsic qualities, 349 organization of, 303-90
involvement, grades of, 253 laboratory, 49, 51, 54
Lack, David, 51
Jeffrey, Richard, 220n, 291n Lakatos, Imre, 8n, 13n, 18n, 86, 89, 112n,
Jenkin, Fleeming, 33n, 38n, 42, 66, 205 115n, 178n, 197, 197n, 220n, 241n
jumping at once, 363, 381. See also labor, Lamarck, Chevalier de, 17n
cognitive, division/distribution of Landre, R., 52n
jumping. See Margolis, Howard language, 30, 31, 34, 75-76, 75n, 80, 94,
justification, 183-84, 220n, 263n 103-4, 120, 150, 221, 227n, 238,
context of, 261 241, 256-57, 259, 262, 273. See also
kinds, natural
Kahnemann, Daniel, 292, 345n philosophy of, 76n
Kant, Immanuel, 62n, 84n, 171-73, 183- scientific, 74, 75-76, 346
84n, 389 embedded in natural languages, 61n, 75,
Kepler, Johannes, 67, 68, 208, 210, 227, 87
232, 255 grammar of, 75
Kim, Jaegwon, 9n relations among, 97
kinds, natural, 80, 95, 96n, 150, 152, 169, Latour, Bruno, 4n, 127n, 141n, 161, 165-
172 67, 168n, 182-86, 198, 303n, 316,
Kirchhoff, Gustav Robert, 145 326, 388n
Kirwan, Richard, 273n, 275-76, 278, 282- Laudan, Larry, 8n, 18n, 86, 89, 94, 110n,
90 112-13n, 115n, 122n, 127n, 130n,
Kitcher, Patricia, 184n 136-37n, 141-50, 157-60, 178n, 197,
knowledge, 170-71n, 188, 219, 234, 295, 204n, 220, 247n, 249n, 262-63, 294n
303n Laudan, Rachel, 110n, 137n
background, 247 Lavoisier, Antoine Laurent, 97-98, 103,
coherentist theories of, 219n 105, 116-17, 132-33, 137, 177, 192,
foundationalist theories of, 162n, 219n 197, 224, 227, 251n, 260, 272-90,
growth of scientific, 63n, 65, 70-71, 293-94, 296, 298, 344-45
89, 127-28, 171-72, 220, 305-6, Legend, 3, 5, 11, 61-62, 64n, 68-71, 89,
388 130n, 272, 275, 288, 294, 327, 388-
historical, 188 89, 390
human, 219-20, 303 Leplin, Jarrett, 127n
mathematical, 170n, 220n, 241n Lesage, George, 158n
perceptual, 170, 224 Levi, Isaac, 94, 197, 220n
Kohn, David, 18n Levins, Richard, 12n
Kordig, Carl, 98n Lewis, David, 170
Kornblith, Hilary, 9n, 162n, 184n Lewontin, Richard, 4n, 12n, 19, 24n, 29,
Kosslyn, Stephen, 63n 30n, 33n, 45n, 80n, 192n
Kottler, Malcolm, 20n, 30n life, cognitive, 61-63, 69, 74, 81, 89
Kripke, Saul, 76, 77n, 79n Lloyd, Elizabeth, 18n, 45n
414 Index
practices (continued) problems, 13, 31, 34, 70, 86n, 235, 262-
sequences of, 87, 91, 303 63, 266, 305
individual, 31, 41, 50-51, 59, 74-75df Barrier, 264df, 268n
82n, 142, 162-65, 179-81, 189, 198, conceptual, 113n
200-201, 208, 220, 221df, 222, 238, Disconnected Range, 264df, 270
298, 353, 256, 260n, 263n, 303. See Displaceable Organism, 264df
also commitments Neighborhood, 264df, 268
two levels of, 89 open, 260df
modification/correction of, 89, 188, 198, qua, 79n
242, 248, 256, 261, 299, 302, 301-3, procedures, 68, 308
306, 334-37, 389 inferential, 85
prior, 244-48, 257 processes, 64, 218n, 231, 275n, 290, 300,
schemata of, 106, 113 303
scientific, 4, 6, 59, 61, 82, 122-24, 131, cognitive/psychological, 68, 79, 79n, 81,
136, 163, 172, 187, 198-200, 202, 184n, 185-194, 198-200, 203-5, 217,
205, 217, 234, 247, 249, 263, 336 218n, 257, 261
sequence of, 74, 89, 92, 173n, 188 adequate, 206df, 207n
tensions among components of, 86 matching. See truth, correspondence theory
total, 124 of
ur-, 299 runaway, 323-24, 341n, 349
variation of, 59n, 61 progress, 10, 91-93, 96n, 111, 117, 123-25,
pre-Copernicans, 96 142n, 152, 162, 180, 186, 306. See
precision, 305 also practices, consensus
predicament, 249, 252, 257, 270 accepted statments, 120
Duhemian, 249-50df. See also Duhem, cognitive, 92-95, 125n, 182, 185-88,
Pierre 191, 197-98, 201, 204, 205, 207,
scientific entrepreneur's, 311df, 329 209, 218n, 221
prediction, 132, 144, 246, 250-58 conceptual, 95, 96, 98, 103-5, 114, 120-
counterfactual, 167 21, 122, 124, 125, 169
successful, 143, 148 cumulative, 110-11
pressure erotetic, 57, 114-15, 117, 120, 124, 125
environmental, 23 experimental, 117, 123
selective, 35 explanatory, 95, 105-112, 111df, 116,
prestige, 315, 318-22, 330-33 117n, 120, 122n, 124, 125, 169, 174
Priestley, Joseph, 97-98, 100-l01n, 102-3, scientific, 4-6, 9, 61, 84, 111n, 117,
105, 114, 116, 117, 132, 137, 148, 127-30, 141n, 151, 187, 297-98, 345
197, 224, 225, 272, 273n, 275-78, progressiveness, 104-5, 116n, 117, 123, 124
280, 283, 288-90 projectibility, 235, 238, 239n, 240n, 251,
priming. See Margolis, Howard 258
principle projects, 72, 73, 82n, 304, 306, 344n
balance sheet, 276n individual epistemic, 73, 93n
charity, 101n proles, 383df, 383-86
humanity, 101n propensities, 65, 68, 69n, 70-71, 78, 79,
logical equivalence, 242-43 85, 198, 224, 226, 228, 238
unification, 172, 173n, 255, 257, 265, cognitive, 179, 187-88
300 eliminative, 237-46
well-adaptedness, 265df, 265-67 generalizing, 241
probabilities, 73, 252, 258n, 335n, 353, higher-order, 191, 195
356n, 358, 374, 387. See also inductive, 234, 235df, 236-37, 241n
Bayes(ian)(ism) inferential, 65, 66, 67, 69-70, 81, 85, 86,
conditions on, 308n 181-82, 191, 221
maximal, 361-64 propositions. See also statements
posterior, 291 incompatible, 68
prior, 68n, 69, 291-93n, 335n storage and retrieval of, 62-63, 70, 78
problem-solving, 63, 70-71, 146, 147n, 346 Provine, William, 49n, 58n
Index 417
research, 86, 89, 112n, 113, 304-5, 316, Schulze, Annedore, 315n
383 science fiction, philosophy of, 153
resources, 69n, 150-51, 190, 275, 308-16, science, 101n, 111n, 112n, 118n, 169n,
329, 333, 383 217n
result, 248, 254 canons of, 3, 7, 34, 86, 158n
revisability, 248 cognitive, 6, 8, 390
revision, 82 history of, 4, 7, 58, 61, 84, 101n, 105-6,
revolution 118, 138n, 149, 158n, 160, 187, 190,
Copernican, 105 198, 202, 224, 248n, 252
scientific, 248, 272 normal, 110, 113, 115n, 248n, 262. See
Rheticus, 206, 210 also Kuhn, Thomas
Roentgen, Wilhelm Konrad, 167n philosophy of, 4, 8-10, 62n, 130n, 150,
Roffeni, Antonio, 231-32 178, 179n, 180, 195, 205, 261
Rohlf, James, 187 rationality of, 6-10, 178, 180-83, 187,
Rosenkrantz, R. D., 258n, 291 192, 196, 202, 291, 293
Ross, Lee, 292 realist view of, 127, 133n
Roughgarden, Jonathan, 120n transformations in, 84
Rouse, Joseph, 303n scientism, 391
Rudwick, Martin, 13n, 15n, 20n, 211-18, scientists, 62, 65, 69-75, 73, 74, 80, 84,
254n 85, 87, 104, 106, 113, 122, 178,
rules, 85, 184, 251, 254 198, 217n, 247, 261, 388. See also
decision, 308, 318, 334, 339 aristos; proles
methodological, 7, 85-86, 178 Secord, James, 34n
straight, 235df, 236 Sedgwick, Adam, 205, 213
Ruse, Michael, 15n, 17n, 18n, 300 selection, 43, 44n, 51-52, 54, 57, 115,
Russell, Bertrand, 5, 389 121n, 300
Russell, E. B., 15 advantages. See reproduction, advantages
artificial, 35, 38, 39
Saint-Hilaire, Geoffroy, 14, 17, 29n Darwinian, 19, 23-25n
salience heuristic. See Tversky, Amos; individual, constrained, 29
Kahneman, Daniel individual, simple, 28-29, 37n, 46, 110
Salmon, Wesley, 106n, 170-72n, 179, 202n, natural, 18, 23n, 24n, 25-27, 32n, 48n,
293n, 299n 50n, 68, 86, 115, 120-21, 300
Sarkar, Husain, 9n Neo-Darwinian, 46-56
Saunders, Peter, 12n, 33n strength of, 48, 49n
Sayre, Anne, 346n self-confidence, 321, 330, 333
scenarios, evolutionary, 70 self-correction, 300, 306, 342
Schaffer, Simon, 169n, 198, 294-97 Sellars, Wilfrid, 67, 162n, 172, 227n
Schaffner, Kenneth, 8n, 141n, 147n, 148n semantics, 75, 131n, 275
Scheffler, Israel, 98n, 178n sense, Fregean, 75-78, 79n
Schelling, Thomas, 73 sentences, 78-79, 82-83, 121, 269-70
schemata, 22, 31-33, 39n, 44-46n, 53, 56, Shapere, Dudley, 8n, 18n, 89, 98n, 220,
65, 74, 82-83, 96n, 106-17, 121, 226-27n, 247n, 249n, 294n
127, 129, 138-39, 143, 145-51, 158, Shapin, Steven, 8n, 127n, 141n, 169n, 198,
160, 171-72, 174, 177, 188, 194, 294-97, 303-4n, 315n, 320n
221, 238, 242, 244-45, 250-51, 256- Shepard, R., 63n
62, 273, 297, 299, 346. See also pat- Sheppard, P. M., 54n
terns significance, 80, 94-95, 100n, 112-14, 118-
accepted, 113, 115n, 144 20, 251, 256, 261-62. See also ques-
correct(ed), 109, 111df, 112, 113-16, tions
117n, 298 cognitive, 5n, 119, 159
explanatory. See also practices derivative, 112, 115
extensions of, 110, 111-12, 113 scientific, 118-19, 252
Schlick, Moritz, 162n significance, cognitive, 5n, 151
Schofield, Robert, 115n, 273n, 276n Silurian period, 58n
Index 419
Watson, James, 4n, 109n, 346n, 352 Woodward, James, 171n, 222
Wayne, Andrew, 241n Woolfenden, G., 119
Weaver, Thomas, 213-14, 217-18 Woolgar, Steve, 141n, 161, 165-67, 168n,
Weber, J., 233n 316
Wegener, Alfred Lothar, 192 world, 61, 76, 95, 106, 122, 127, 128-29,
Weinberg, Wilhelm, 42n, 43, 44n 131, 157, 171, 181, 236n, 241,
well-being, cognitive, 72 390
Westfall, R. S., 183n, 230n independent of human cognition, 127. See
Westman, Robert, 208n, 209n also realism
Whewell, William, 389 structure of, 94, 153
whole theories, 118, 129, 143, 150n Worrall, John, 86n, 123n, 144n, 179n, 197,
Wilberforce, Samuel, 180-82 247n
Williams, David, 213-14, 217-18 Wright, Sewall, 43, 44n, 45n, 46n, 47-50,
Williams, George, 4n, 52 55, 58n, 83, 113
Wilson, David S., 120n
Wilson, E. O., 70, 154
Wittgenstein, Ludwig, 8n, 184n, 295n Yule, George Udney, 42n
Wolf, Emil, 144n, 145
Wollaston, Thomas Vernon, 36, 38, 39, 40,
42 Zahar, Elie, 197n, 247n